AI – Page 5

LG TVs’ unremovable Copilot shortcut is the least of smart TVs’ AI problems

AI, chatbots, copilot, generative ai, lg, microsoft, samsung, Tech, TVs / Tim Belzer / December 19, 2025

But Copilot will still be integrated into Tizen OS, and Samsung appears eager to push chatbots into TVs, including by launching Perplexity’s first TV app. Amazon, which released Fire TVs with Alexa+ this year, is also exploring putting chatbots into TVs.

After the backlash LG faced this week, companies may reconsider installing AI apps on people’s smart TVs. A better use of large language models in TVs may be as behind-the-scenes tools to improve TV watching. People generally don’t buy smart TVs to make it easier to access chatbots.

But this development is still troubling for anyone who doesn’t want an AI chatbot in their TV at all.

Some people don’t want chatbots in their TVs

Subtle integrations of generative AI that make it easier for people to do things like figure out the name of “that movie” may have practical use, but there are reasons to be wary of chatbot-wielding TVs.

Chatbots add another layer of complexity to understanding how a TV tracks user activity. With a chatbot involved, smart TV owners will be subject to complicated smart TV privacy policies and terms of service, as well as the similarly verbose rules of third-party AI companies. This will make it harder for people to understand what data they’re sharing with companies, and there’s already serious concern about the boundaries smart TVs are pushing to track users, including without consent.

Chatbots can also contribute to smart TV bloatware. Unwanted fluff, like games, shopping shortcuts, and flashy ads, already disrupts people who just want to watch TV.

LG’s Copilot web app is worthy of some grousing, but not necessarily because of the icon that users will eventually be able to delete. The more pressing issue is the TV industry’s shift toward monetizing software with user tracking and ads.

If you haven’t already, now is a good time to check out our guide to breaking free from smart TV ads and tracking.

LG TVs’ unremovable Copilot shortcut is the least of smart TVs’ AI problems Read More »

We asked four AI coding agents to rebuild Minesweeper—the results were explosive

Agent, AI, AI gaming, Claude Code, coding, gaming, Gemini, test / Rejus Almole / December 19, 2025

How do four modern LLMs do at re-creating a simple Windows gaming classic?

Which mines are mine, and which are AI? Credit: Aurich Lawson | Getty Images

The idea of using AI to help with computer programming has become a contentious issue. On the one hand, coding agents can make horrific mistakes that require a lot of inefficient human oversight to fix, leading many developers to lose trust in the concept altogether. On the other hand, some coders insist that AI coding agents can be powerful tools and that frontier models are quickly getting better at coding in ways that overcome some of the common problems of the past.

To see how effective these modern AI coding tools are becoming, we decided to test four major models with a simple task: re-creating the classic Windows game Minesweeper. Since it’s relatively easy for pattern-matching systems like LLMs to play off of existing code to re-create famous games, we added in one novelty curveball as well.

Our straightforward prompt:

Make a full-featured web version of Minesweeper with sound effects that

1) Replicates the standard Windows game and

2) implements a surprise, fun gameplay feature.

Include mobile touchscreen support.

Ars Senior AI Editor Benj Edwards fed this task into four AI coding agents with terminal (command line) apps: OpenAI’s Codex based on GPT-5, Anthropic’s Claude Code with Opus 4.5, Google’s Gemini CLI, and Mistral Vibe. The agents then directly manipulated HTML and scripting files on a local machine, guided by a “supervising” AI model that interpreted the prompt and assigned coding tasks to parallel LLMs that can use software tools to execute the instructions. All AI plans were paid for privately with no special or privileged access given by the companies involved, and the companies were unaware of these tests taking place.

Ars Senior Gaming Editor (and Minesweeper expert) Kyle Orland then judged each example blind, without knowing which model generated which Minesweeper clone. Those somewhat subjective and non-rigorous results are below.

For this test, we used each AI model’s unmodified code in a “single shot” result to see how well these tools perform without any human debugging. In the real world, most sufficiently complex AI-generated code would go through at least some level of review and tweaking by a human software engineer who could spot problems and address inefficiencies.

We chose this test as a sort of simple middle ground for the current state of AI coding. Cloning Minesweeper isn’t a trivial task that can be done in just a handful of lines of code, but it’s also not an incredibly complex system that requires many interlocking moving parts.

Minesweeper is also a well-known game, with many versions documented across the Internet. That should give these AI agents plenty of raw material to work from and should be easier for us to evaluate than a completely novel program idea. At the same time, our open-ended request for a new “fun” feature helps demonstrate each agent’s penchant for unguided coding “creativity,” as well as their ability to create new features on top of an established game concept.

With all that throat-clearing out of the way, here’s our evaluation of the AI-generated Minesweeper clones, complete with links that you can use to play them yourselves.

Agent 1: Mistral Vibe

Play it for yourself

Just ignore that Custom button. It’s purely for show. Credit: Benj Edwards

Implementation

Right away, this version loses points for not implementing chording—the technique that advanced Minesweeper players use to quickly clear all the remaining spaces surrounding a number that already has sufficient flagged mines. Without this feature, this version feels more than a little clunky to play.

I’m also a bit perplexed by the inclusion of a “Custom” difficulty button that doesn’t seem to do anything. It’s like the model realized that customized board sizes were a thing in Minesweeper but couldn’t figure out how to implement this relatively basic feature.

The game works fine on mobile, but marking a square with a flag requires a tricky long-press on a tiny square that also triggers selector handles that are difficult to clear. So it’s not an ideal mobile interface.

Presentation

This was the only working version we tested that didn’t include sound effects. That’s fair, since the original Windows Minesweeper also didn’t include sound, but it’s still a notable relative omission since the prompt specifically asked for it.

The all-black “smiley face” button to start a game is a little off-putting, too, compared to the bright yellow version that’s familiar to both Minesweeper players and emoji users worldwide. And while that smiley face does start a new game when clicked, there’s also a superfluous “New Game” button taking up space for some reason.

“Fun” feature

The closest thing I found to a “fun” new feature here was the game adding a rainbow background pattern on the grid when I completed a game. While that does add a bit of whimsy to a successful game, I expected a little more.

Coding experience

Benj notes that he was pleasantly surprised by how well Mistral Vibe performed as an open-weight model despite lacking the big-money backing of the other contenders. It was relatively slow, however (third fastest out of four), and the result wasn’t great. Ultimately, its performance so far suggests that with more time and more training, a very capable AI coding agent may eventually emerge.

Overall rating: 4/10

This version got many of the basics right but left out chording and didn’t perform well on the small presentational and “fun” touches.

Agent 2: OpenAI Codex

Play it for yourself

I can’t tell you how much I appreciate those chording instructions at the bottom. Credit: Benj Edwards

Implementation

Not only did this agent include the crucial “chording” feature, but it also included on-screen instructions for using it on both PC and mobile browsers. I was further impressed by the option to cycle through “?” marks when marking squares with flags, an esoteric feature I feel even most human Minesweeper cloners might miss.

On mobile, the option to hold your finger down on a square to mark a flag is a nice touch that makes this the most enjoyable handheld version we tested.

Presentation

The old-school emoticon smiley-face button is pretty endearing, especially when you blow up and get a red-tinted “X(“. I was less impressed by the playfield “graphics,” which use a simple “*” for revealed mines and an ugly red “F” for flagged tiles.

The beeps-and-boops sound effects reminded me of my first old-school, pre-Sound-Blaster PC from the late ’80s. That’s generally a good thing, but I still appreciated the game giving me the option to turn them off.

“Fun” feature

The “Surprise: Lucky Sweep Bonus” listed in the corner of the UI explains that clicking the button gives you a free safe tile when available. This can be pretty useful in situations where you’d otherwise be forced to guess between two tiles that are equally likely to be mines.

Overall, though, I found it a bit odd that the game gives you this bonus only after you find a large, cascading field of safe tiles with a single click. It mostly functions as a “win more” button rather than a feature that offers a good balance of risk versus reward.

Coding experience

OpenAI Codex has a nice terminal interface with features similar to Claude Code (local commands, permission management, and interesting animations showing progress), and it’s fairly pleasant to use (OpenAI also offers Codex through a web interface, but we did not use that for this evaluation). However, Codex took roughly twice as long to code a functional game than Claude Code did, which might contribute to the strong result here.

Overall: 9/10

The implementation of chording and cute presentation touches push this to the top of the list. We just wish the “fun” feature was a bit more fun.

Agent 3: Anthropic Claude Code

Play it for yourself

The Power Mod powers on display here make even Expert boards pretty trivial to complete. Credit: Benj Edwards

Implementation

Once again, we get a version that gets all the gameplay basics right but is missing the crucial chording feature that makes truly efficient Minesweeper play possible. This is like playing Super Mario Bros. without the run button or Ocarina of Time without Z-targeting. In a word: unacceptable.

The “flag mode” toggle on the mobile version of this game is perfectly functional, but it’s a little clunky to use. It also visually cuts off a portion of the board at the larger game sizes.

Presentation

Presentation-wise, this is probably the most polished version we tested. From the use of cute emojis for the “face” button to nice-looking bomb and flag graphics and simple but effective sound effects, this looks more professional than the other versions we tested.

That said, there are some weird presentation issues. The “beginner” grid has weird gaps between columns, for instance. The borders of each square and the flag graphics can also become oddly grayed out at points, especially when using Power Mode (see below).

“Fun” feature

The prominent “Power Mode” button in the lower-right corner offers some pretty fun power-ups that alter the core Minesweeper formula in interesting ways. But the actual powers are a bit hit-and-miss.

I especially liked the “Shield” power, which protects you from an errant guess, and the “Blast” power, which seems to guarantee a large cascade of revealed tiles wherever you click. But the “X-Ray” power, which reveals every bomb for a few seconds, could be easily exploited by a quick player (or a crafty screenshot). And the “Freeze” power is rather boring, just stopping the clock for a few seconds and amounting to a bit of extra time.

Overall, the game hands out these new powers like candy, which makes even an Expert-level board relatively trivial with Power Mode active. Simply choosing “Power Mode” also seems to mark a few safe squares right after you start a game, making things even easier. So while these powers can be “fun,” they also don’t feel especially well-balanced.

Coding experience

Of the four tested models, Claude Code with Opus 4.5 featured the most pleasant terminal interface experience and the fastest overall coding experience (Claude Code can also use Sonnet 4.5, which is even faster, but the results aren’t quite as full-featured in our experience). While we didn’t precisely time each model, Opus 4.5 produced a working Minesweeper in under five minutes. Codex took at least twice as long, if not longer, while Mistral took roughly three or four times as long as Claude Code. Gemini, meanwhile, took hours of tinkering to get two non-working results.

Overall: 7/10

The lack of chording is a big omission, but the strong presentation and Power Mode options give this effort a passable final score.

Agent 4: Google Gemini CLI

Play it for yourself

So… where’s the game? Credit: Benj Edwards

Implementation, presentation, etc.

Gemini CLI did give us a few gray boxes you can click, but the playfields are missing. While interactive troubleshooting with the agent may have fixed the issue, as a “one-shot” test, the model completely failed.

Coding experience

Of the four coding agents we tested, Gemini CLI gave Benj the most trouble. After developing a plan, it was very, very slow at generating any usable code (about an hour per attempt). The model seemed to get hung up attempting to manually create WAV file sound effects and insisted on requiring React external libraries and a few other overcomplicated dependencies. The result simply did not work.

Benj actually bent the rules and gave Gemini a second chance, specifying that the game should use HTML5. When the model started writing code again, it also got hung up trying to make sound effects. Benj suggested using the WebAudio framework (which the other AI coding agents seemed to be able to use), but the result didn’t work, which you can see at the link above.

Unlike the other models tested, Gemini CLI apparently uses a hybrid system of three different LLMs for different tasks (Gemini 2.5 Flash Lite, 2.5 Flash, and 2.5 Pro were available at the level of the Google account Benj paid for). When you’ve completed your coding session and quit the CLI interface, it gives you a readout of which model did what.

In this case, it didn’t matter because the results didn’t work. But it’s worth noting that Gemini 3 coding models are available for other subscription plans that were not tested here. For that reason, this portion of the test could be considered “incomplete” for Google CLI.

Overall: 0/10 (Incomplete)

Final verdict

OpenAI Codex wins this one on points, in no small part because it was the only model to include chording as a gameplay option. But Claude Code also distinguished itself with strong presentational flourishes and quick generation time. Mistral Vibe was a significant step down, and Google CLI based on Gemini 2.5 was a complete failure on our one-shot test.

While experienced coders can definitely get better results via an interactive, back-and-forth code editing conversation with an agent, these results show how capable some of these models can be, even with a very short prompt on a relatively straightforward task. Still, we feel that our overall experience with coding agents on other projects (more on that in a future article) generally reinforces the idea that they currently function best as interactive tools that augment human skill rather than replace it.

Kyle Orland has been the Senior Gaming Editor at Ars Technica since 2012, writing primarily about the business, tech, and culture behind video games. He has journalism and computer science degrees from University of Maryland. He once wrote a whole book about Minesweeper.

We asked four AI coding agents to rebuild Minesweeper—the results were explosive Read More »

YouTube bans two popular channels that created fake AI movie trailers

AI, Ai video, Artificial Intelligence, Google, Tech, youtube / Tim Belzer / December 18, 2025

Deadline reports that the behavior of these creators ran afoul of YouTube’s spam and misleading-metadata policies. At the same time, Google loves generative AI—YouTube has added more ways for creators to use generative AI, and the company says more gen AI tools are coming in the future. It’s quite a tightrope for Google to walk.

AI movie trailers — A selection of videos from the now-defunct Screen Culture channel. Credit: Ryan Whitwam

While passing off AI videos as authentic movie trailers is definitely spammy conduct, the recent changes to the legal landscape could be a factor, too. Disney recently entered into a partnership with OpenAI, bringing its massive library of characters to the company’s Sora AI video app. At the same time, Disney sent a cease-and-desist letter to Google demanding the removal of Disney content from Google AI. The letter specifically cited AI content on YouTube as a concern.

Both the banned trailer channels made heavy use of Disney properties, sometimes even incorporating snippets of real trailers. For example, Screen Culture created 23 AI trailers for The Fantastic Four: First Steps, some of which outranked the official trailer in searches. It’s unclear if either account used Google’s Veo models to create the trailers, but Google’s AI will recreate Disney characters without issue.

While Screen Culture and KH Studio were the largest purveyors of AI movie trailers, they are far from alone. There are others with five and six-digit subscriber counts, some of which include disclosures about fan-made content. Is that enough to save them from the ban hammer? Many YouTube viewers probably hope not.

YouTube bans two popular channels that created fake AI movie trailers Read More »

School security AI flagged clarinet as a gun. Exec says it wasn’t an error.

AI, ai gun detection, AI security, Artificial Intelligence, Policy, school shooting, zeroeyes / Paul Patrick / December 18, 2025

Human review didn’t stop AI from triggering lockdown at panicked middle school.

A Florida middle school was locked down last week after an AI security system called ZeroEyes mistook a clarinet for a gun, reviving criticism that AI may not be worth the high price schools pay for peace of mind.

Human review of the AI-generated false flag did not stop police from rushing to Lawton Chiles Middle School. Cops expected to find “a man in the building, dressed in camouflage with a ‘suspected weapon pointed down the hallway, being held in the position of a shouldered rifle,’” a Washington Post review of the police report said.

Instead, after finding no evidence of a shooter, cops double-checked with dispatchers who confirmed that a closer look at the images indicated that “the suspected rifle might have been a band instrument.” Among panicked students hiding in the band room, police eventually found the suspect, a student “dressed as a military character from the Christmas movie Red One for the school’s Christmas-themed dress-up day,” the Post reported.

ZeroEyes cofounder Sam Alaimo told the Post that the AI performed exactly as it should have in this case, adopting a “better safe than sorry” outlook. A ZeroEyes spokesperson told Ars that “school resource officers, security directors and superintendents consistently ask us to be proactive and forward them an alert if there is any fraction of a doubt that the threat might be real.”

“We don’t think we made an error, nor does the school,” Alaimo said. “That was better to dispatch [police] than not dispatch.”

Cops left after the confused student confirmed he was “unaware” that the way he was holding his clarinet could have triggered that alert, the Post reported. But ZeroEyes’ spokesperson claimed he was “intentionally holding the instrument in the position of a shouldered rifle.” And seemingly rather than probe why the images weren’t more carefully reviewed to prevent a false alarm on campus, the school appeared to agree with ZeroEyes and blame the student.

“We did not make an error, and the school was pleased with the detection and their response,” ZeroEyes’ spokesperson said.

School warns students not to trigger AI

In a letter to parents, the principal, Melissa Laudani, reportedly told parents that “while there was no threat to campus, I’d like to ask you to speak with your student about the dangers of pretending to have a weapon on a school campus.” Along similar lines, Seminole County Public Schools (SCPS) communications officer, Katherine Crnkovich, emphasized in an email to Ars to “please make sure it is noted that this student wasn’t simply carrying a clarinet. This individual was holding it as if it were a weapon.”

However, warning students against brandishing ordinary objects like weapons isn’t a perfect solution. Video footage from a Texas high school in 2023 showed that ZeroEyes can sometimes confuse shadows for guns, accidentally flagging a student simply walking into school as a potential threat. The advice also ignores that ZeroEyes last year reportedly triggered a lockdown and police response after detecting two theater kids using prop guns to rehearse a play. And a similar AI tool called Omnilert made national headlines confusing an empty Doritos bag with a gun, which led to a 14-year-old Baltimore sophomore’s arrest. In that case, the student told the American Civil Liberties Union that he was just holding the chips when AI sent “like eight cop cars” to detain him.

For years, school safety experts have warned that AI tools like ZeroEyes take up substantial resources even though they are “unproven,” the Post reported. ZeroEyes’ spokesperson told Ars that “in most cases, ZeroEyes customers will never receive a ‘false positive,’” but the company is not transparent about how many false positives it receives or how many guns have been detected. An FAQ only notes that “we are always looking to minimize false positives and are constantly improving our learning models based on data collected.” In March, as some students began questioning ZeroEyes after it flagged a Nerf gun at a Pennsylvania university, a nearby K-12 private school, Germantown Academy, confirmed that its “system often makes ‘non-lethal’ detections.”

One critic, school safety consultant Kenneth Trump, suggested in October that these tools are “security theater,” with firms like ZeroEyes lobbying for taxpayer dollars by relying on what the ACLU called “misleading” marketing to convince schools that tools are proactive solutions to school shootings. Seemingly in response to this backlash, StateScoop reported that days after it began probing ZeroEyes in 2024, the company scrubbed a claim from its FAQ that said ZeroEyes “can prevent active shooter and mass shooting incidents.”

At Lawton Chiles Middle School, “the children were never in any danger,” police confirmed, but experts question if false positives cause students undue stress and suspicion, perhaps doing more harm than good in absence of efficacy studies. Schools may be better off dedicating resources to mental health services proven to benefit kids, some critics have suggested.

Laudani’s letter encouraged parents to submit any questions they have about the incident, but it’s hard to gauge if anyone’s upset. Asked if parents were concerned or if ZeroEyes has ever triggered lockdown at other SCPS schools, Crnkovich told Ars that SCPS does not “provide details regarding the specific school safety systems we utilize.”

It’s clear, however, that SCPS hopes to expand its use of ZeroEyes. In November, Florida state Senator Keith Truenow submitted a request to install “significantly more cameras”—about 850—equipped with ZeroEyes across the school district. Truenow backed up his request for $500,000 in funding over the next year by claiming that “the more [ZeroEyes] coverage there is, the more protected students will be from potential gun violence.”

AI false alarms pose dangers to students

ZeroEyes is among the most popular tools attracting heavy investments from schools in 48 states, which hope that AI gun detection will help prevent school shootings. The AI technology is embedded in security cameras, trained on images of people holding guns, and can supposedly “detect as little as an eighth of an inch of a gun,” an ABC affiliate in New York reported.

Monitoring these systems continually, humans review AI flags, then text any concerning images detected to school superintendents. Police are alerted when human review determines images may constitute actual threats. ZeroEyes’ spokesperson told Ars that “it has detected more than 1,000 weapons in the last three years.” Perhaps most notably, ZeroEyes “detected a minor armed with an AK-47 rifle on an elementary school campus in Texas,” where no shots were fired, StateScoop reported last year.

Schools invest tens or, as the SCPS case shows, even hundreds of thousands annually, the exact amount depending on the number of cameras they want to employ and other variables impacting pricing. ZeroEyes estimates that most schools pay $60 per camera monthly. Bigger contracts can discount costs. In Kansas, a statewide initiative equipping 25 cameras at 1,300 schools with ZeroEyes was reportedly estimated to cost $8.5 million annually. Doubling the number of cameras didn’t provide much savings, though, with ZeroEyes looking to charge $15.2 million annually to expand coverage.

To critics, it appears that ZeroEyes is attempting to corner the market on AI school security, standing to profit off schools’ fears of shootings, while showing little proof of the true value of its systems. Last year, ZeroEyes reported its revenue grew 300 percent year over year from 2023 to 2024, after assisting in “more than ten arrests through its thousands of detections, verifications, and notifications to end users and law enforcement.”

Curt Lavarello, the executive director of the School Safety Advocacy Council, told the ABC News affiliate that “all of this technology is very, very expensive,” considering that “a lot of products … may not necessarily do what they’re being sold to do.”

Another problem, according to experts who have responded to some of the country’s deadliest school shootings, is that while ZeroEyes’ human reviewers can alert police in “seconds,” police response can often take “several minutes.” That delay could diminish ZeroEyes’ impact, one expert suggested, noting that at an Oregon school he responded to, there was a shooter who “shot 25 people in 60 seconds,” StateScoop reported.

In Seminole County, where the clarinet incident happened, ZeroEyes has been used since 2021, but SCPS would not confirm if any guns have ever been detected to justify next year’s desired expansion. It’s possible that SCPS has this information, as Sen. Truenow noted in his funding request that ZeroEyes can share reports with schools “to measure the effectiveness of the ZeroEyes deployment” by reporting on “how many guns were detected and alerted on campus.”

ZeroEyes’ spokesperson told Ars that “trained former law enforcement and military make split-second, life-or-death decisions about whether the threat is real,” which is supposed to help reduce false positives that could become more common as SCPS adds ZeroEyes to many more cameras.

Amanda Klinger, the director of operations at the Educator’s School Safety Network, told the Post that too many false alarms could carry two risks. First, more students could be put in dangerous situations when police descend on schools where they anticipate confronting an active shooter. And second, cops may become fatigued by false alarms, perhaps failing to respond with urgency over time. For students, when AI labels them as suspects, it can also be invasive and humiliating, reports noted.

“We have to be really clear-eyed about what are the limitations of these technologies,” Klinger said.

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

School security AI flagged clarinet as a gun. Exec says it wasn’t an error. Read More »

LLMs’ impact on science: Booming publications, stagnating quality

AI, LLMs, Science, scientific publications / Kris Guyer / December 18, 2025

This effect was likely to be most pronounced in people that weren’t native speakers of English. If the researchers limited the analysis to people with Asian names working at institutions in Asia, their rate of submissions to bioRxiv and SSRN nearly doubled once they started using AI and rose by over 40 percent at the arXiv. This suggests that people who may not have the strongest English skills are using LLMs to overcome a major bottleneck: producing compelling text.

Quantity vs. quality

The value of producing compelling text should not be underestimated. “Papers with clear but complex language are perceived to be stronger and are cited more frequently,” the researchers note, suggesting that we may use the quality of writing as a proxy for the quality of the research it’s describing. And they found some indication of that here, as non-LLM-assisted papers were more likely to be published in the peer reviewed literature if they used complex language (the abstracts were scored for language complexity using a couple of standard measures).

But the dynamic was completely different for LLM-produced papers. The complexity of language in papers written with an LLM was generally higher than for those using natural language. But they were less likely to end up being published. “For LLM-assisted manuscripts,” the researchers write, “the positive correlation between linguistic complexity and scientific merit not only disappears, it inverts.”

But not all of the differences were bleak. When the researchers checked the references being used in AI-assisted papers, they found that the LLMs weren’t just citing the same papers that everyone else did. They instead cited a broader range of sources, and were more likely to cite books and recent papers. So, there’s a chance that AI use could ultimately diversify the published research that other researchers consider (assuming they check their own references, which they clearly should).

What does this tell us?

There are a couple of cautions for interpreting these results. One, acknowledged by the researchers, is that people may be using AI to produce initial text that’s then heavily edited, and that may be mislabeled as human-produced text here. So, the overall prevalence of AI use is likely to be higher. The other is that some manuscripts may take a while to get published, so their use of that as a standard for scientific quality may penalize more recent drafts—which are more likely to involve AI use. These may ultimately bias some of the results, but the effects the authors saw were so large that they’re unlikely to go away entirely.

LLMs’ impact on science: Booming publications, stagnating quality Read More »

OpenAI’s new ChatGPT image generator makes faking photos easy

AI, AI image generator, AI image generators, API, Biz & IT, chatgpt, Deepfakes, generative ai, Google, image synthesis, machine learning, multimodal AI, openai / Rejus Almole / December 17, 2025

For most of photography’s roughly 200-year history, altering a photo convincingly required either a darkroom, some Photoshop expertise, or, at minimum, a steady hand with scissors and glue. On Tuesday, OpenAI released a tool that reduces the process to typing a sentence.

It’s not the first company to do so. While OpenAI had a conversational image-editing model in the works since GPT-4o in 2024, Google beat OpenAI to market in March with a public prototype, then refined it to a popular model called Nano Banana image model (and Nano Banana Pro). The enthusiastic response to Google’s image-editing model in the AI community got OpenAI’s attention.

OpenAI’s new GPT Image 1.5 is an AI image synthesis model that reportedly generates images up to four times faster than its predecessor and costs about 20 percent less through the API. The model rolled out to all ChatGPT users on Tuesday and represents another step toward making photorealistic image manipulation a casual process that requires no particular visual skills.

The “Galactic Queen of the Universe” added to a photo of a room with a sofa using GPT Image 1.5 in ChatGPT.

GPT Image 1.5 is notable because it’s a “native multimodal” image model, meaning image generation happens inside the same neural network that processes language prompts. (In contrast, DALL-E 3, an earlier OpenAI image generator previously built into ChatGPT, used a different technique called diffusion to generate images.)

This newer type of model, which we covered in more detail in March, treats images and text as the same kind of thing: chunks of data called “tokens” to be predicted, patterns to be completed. If you upload a photo of your dad and type “put him in a tuxedo at a wedding,” the model processes your words and the image pixels in a unified space, then outputs new pixels the same way it would output the next word in a sentence.

Using this technique, GPT Image 1.5 can more easily alter visual reality than earlier AI image models, changing someone’s pose or position, or rendering a scene from a slightly different angle, with varying degrees of success. It can also remove objects, change visual styles, adjust clothing, and refine specific areas while preserving facial likeness across successive edits. You can converse with the AI model about a photograph, refining and revising, the same way you might workshop a draft of an email in ChatGPT.

OpenAI’s new ChatGPT image generator makes faking photos easy Read More »

Bursting AI bubble may be EU’s “secret weapon” in clash with Trump, expert says

Accenture, AI, Digital Services Act, Donald Trump, elon musk, European Commission, Mistral, Policy, Spotify, trump administration, X / Mike M. / December 17, 2025

Spotify and Accenture caught in crossfire as Trump attacks EU tech regulations.

The US threatened to restrict some of the largest service providers in the European Union as retaliation for EU tech regulations and investigations are increasingly drawing Donald Trump’s ire.

On Tuesday, the Office of the US Trade Representative (USTR) issued a warning on X, naming Spotify, Accenture, Amadeus, Mistral, Publicis, and DHL among nine firms suddenly yanked into the middle of the US-EU tech fight.

“The European Union and certain EU Member States have persisted in a continuing course of discriminatory and harassing lawsuits, taxes, fines, and directives against US service providers,” USTR’s post said.

The clash comes after Elon Musk’s X became the first tech company fined for violating the EU’s Digital Services Act, which is widely considered among the world’s strictest tech regulations. Trump was not appeased by the European Commission (EC) noting that X was not ordered to pay the maximum possible fine. Instead, the $140 million fine sparked backlash within the Trump administration, including from Vice President JD Vance, who slammed the fine as “censorship” of X and its users.

Asked for comment on the USTR’s post, an EC spokesperson told Ars that the EU intends to defend its tech regulations while implementing commitments from a Trump trade deal that the EU struck in August.

“The EU is an open and rules-based market, where companies from all over the world do business successfully and profitably,” the EC’s spokesperson said. “As we have made clear many times, our rules apply equally and fairly to all companies operating in the EU,” ensuring “a safe, fair and level playing field in the EU, in line with the expectations of our citizens. We will continue to enforce our rules fairly, and without discrimination.”

Trump on shaky ground due to “AI bubble”

On X, the USTR account suggested that the EU was overlooking that US companies “provide substantial free services to EU citizens and reliable enterprise services to EU companies,” while supporting “millions of jobs and more than $100 billion in direct investment in Europe.”

To stop what Trump views as “overseas extortion” of American tech companies, the USTR said the US was prepared to go after EU service providers, which “have been able to operate freely in the United States for decades, benefitting from access to our market and consumers on a level playing field.”

“If the EU and EU Member States insist on continuing to restrict, limit, and deter the competitiveness of US service providers through discriminatory means, the United States will have no choice but to begin using every tool at its disposal to counter these unreasonable measures,” USTR’s post said. “Should responsive measures be necessary, US law permits the assessment of fees or restrictions on foreign services, among other actions.”

The pushback comes after the Trump administration released a November national security report that questioned how long the EU could remain a “reliable” ally as overregulation of its tech industry could hobble both its economy and military strength. Claiming that the EU was only “doubling down” on such regulations, the EU “will be unrecognizable in 20 years or less,” the report predicted.

“We want Europe to remain European, to regain its civilizational self-confidence, and to abandon its failed focus on regulatory suffocation,” the report said.

However, the report acknowledged that “Europe remains strategically and culturally vital to the United States.”

“Transatlantic trade remains one of the pillars of the global economy and of American prosperity,” the report said. “European sectors from manufacturing to technology to energy remain among the world’s most robust. Europe is home to cutting-edge scientific research and world-leading cultural institutions. Not only can we not afford to write Europe off—doing so would be self-defeating for what this strategy aims to achieve.”

At least one expert in the EU has suggested that the EU can use this acknowledgement as leverage, while perhaps even using the looming threat of the supposed American “AI bubble” bursting to pressure Trump into backing off EU tech laws.

In an op-ed for The Guardian, Johnny Ryan, the director of Enforce, a unit of the Irish Council for Civil Liberties, suggested that the EU could even throw Trump’s presidency into “crisis” by taking bold steps that Trump may not see coming.

EU can take steps to burst “AI bubble”

According to Ryan, the national security report made clear that the EU must fight the US or else “perish.” However, the EU has two “strong cards” to play if it wants to win the fight, he suggested.

Right now, market analysts are fretting about an “AI bubble,” with US investment in AI far outpacing potential gains until perhaps 2030. A Harvard University business professor focused on helping businesses implement cutting-edge technology like generative AI, Andy Wu, recently explained that AI’s big problem is that “everyone can imagine how useful the technology will be, but no one has figured out yet how to make money.”

“If the market can keep the faith to persist, it buys the necessary time for the technology to mature, for the costs to come down, and for companies to figure out the business model,” Wu said. But US “companies can end up underwater if AI grows fast but less rapidly than they hope for,” he suggested.

During this moment, Ryan wrote, it’s not just AI firms with skin in the game, but potentially all of Trump’s supporters. The US is currently on “shaky economic ground” with AI investment accounting “for virtually all (92 percent) GDP growth in the first half of this year.”

“The US’s bet on AI is now so gigantic that every MAGA voter’s pension is bound to the bubble’s precarious survival,” Ryan said.

Ursula von der Leyen, the president of the European Commission, could exploit this apparent weakness first by messing with one of the biggest players in America’s AI industry, Nvidia, then by ramping up enforcement of the tech laws Trump loathes.

According to Ryan, “Dutch company ASML commands a global monopoly on the microchip-etching machines that use light to carve patterns on silicon,” and Nvidia needs those machines if it wants to remain the world’s most valuable company. Should the US GDP remain reliant on AI investment for growth, von der Leyen could use export curbs on that technology like a “lever,” Ryan said, controlling “whether and by how much the US economy expands or contracts.”

Withholding those machines “would be difficult for Europe” and “extremely painful for the Dutch economy,” Ryan noted, but “it would be far more painful for Trump.”

Another step the EU could take is even “easier,” Ryan suggested. It could go even harder on the enforcement of tech regulations based on evidence of mismanaged data surfaced in lawsuits against giants like Google and Meta. For example, it seems clear that Meta may have violated the EU’s General Data Protection Regulation (GDPR), after the Facebook owner was “unable to tell a US court that what its internal systems do with your data, or who can access it, or for what purpose.”

“This data free-for-all lets big tech companies train their AI models on masses of everyone’s data, but it is illegal in Europe, where companies are required to carefully control and account for how they use personal data,” Ryan wrote. “All Brussels has to do is crack down on Ireland, which for years has been a wild west of lax data enforcement, and the repercussions will be felt far beyond.”

Taking that step would also arguably make it harder for tech companies to secure AI investments, since firms would have to disclose that their “AI tools are barred from accessing Europe’s valuable markets,” Ryan said.

Calling the reaction to the X fine “extreme,” Ryan pushed for von der Leyen to advance on both fronts, forecasting that “the AI bubble would be unlikely to survive this double shock” and likely neither could Trump’s approval ratings. There’s also a possibility that tech firms could pressure Trump to back down if coping with any increased enforcement threatens AI progress.

Although Wu suggested that Big Tech firms like Google and Meta would likely be “insulated” from the AI bubble bursting, Google CEO Sundar Pichai doesn’t seem so sure. In November, Pichai told the BBC that if AI investments didn’t pay off quickly enough, he thinks “no company is going to be immune, including us.”

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

Bursting AI bubble may be EU’s “secret weapon” in clash with Trump, expert says Read More »

Google releases Gemini 3 Flash, promising improved intelligence and efficiency

AI, Artificial Intelligence, Gemini 3, generative ai, Google, search / Rejus Almole / December 17, 2025

Google began its transition to Gemini 3 a few weeks ago with the launch of the Pro model, and the arrival of Gemini 3 Flash kicks it into high gear. The new, faster Gemini 3 model is coming to the Gemini app and search, and developers will be able to access it immediately via the Gemini API, Vertex AI, AI Studio, and Antigravity. Google’s bigger gen AI model is also picking up steam, with both Gemini 3 Pro and its image component (Nano Banana Pro) expanding in search.

This may come as a shock, but Google says Gemini 3 Flash is faster and more capable than its previous base model. As usual, Google has a raft of benchmark numbers that show modest improvements for the new model. It bests the old 2.5 Flash in basic academic and reasoning tests like GPQA Diamond and MMMU Pro (where it even beats 3 Pro). It gets a larger boost in Humanity’s Last Exam (HLE), which tests advanced domain-specific knowledge. Gemini 3 Flash has tripled the old models’ score in HLE, landing at 33.7 percent without tool use. That’s just a few points behind the Gemini 3 Pro model.

Google is talking up Gemini 3 Flash’s coding skills, and the provided benchmarks seem to back that talk up. Over the past year, Google has mostly pushed its Pro models as the best for generating code, but 3 Flash has done a lot of catching up. In the popular SWE-Bench Verified test, Gemini 3 Flash has gained almost 20 points on the 2.5 branch.

The new model is also a lot less likely to get general-knowledge questions wrong. In the Simple QA Verified test, Gemini 3 Flash scored 68.7 percent, which is only a little below Gemini 3 Pro. The last Flash model scored just 28.1 percent on that test. At least as far as the evaluation scores go, Gemini 3 Flash performs much closer to Google’s Pro model versus the older 2.5 family. At the same time, it’s considerably more efficient, according to Google.

One of Gemini 3 Pro’s defining advances was its ability to generate interactive simulations and multimodal content. Gemini 3 Flash reportedly retains that underlying capability. Gemini 3 Flash offers better performance than Gemini 2.5 Pro did, but it runs workloads three times faster. It’s also a lot cheaper than the Pro models if you’re paying per token. One million input tokens for 3 Flash will run devs $0.50, and a million output tokens will cost $3. However, that’s an increase compared to Gemini 2.5 Flash input and output at $0.30 and $2.50, respectively. The Pro model’s tokens are $2 (1M input) and $12 (1M output).

Google releases Gemini 3 Flash, promising improved intelligence and efficiency Read More »

Browser extensions with 8 million users collect extended AI conversations

AI, ai bot, Biz & IT, browser extensions, chrome, EDGE, privacy, Security / Kris Guyer / December 17, 2025

Besides ChatGPT, Claude, and Gemini, the extensions harvest all conversations from Copilot, Perplexity, DeepSeek, Grok, and Meta AI. Koi said the full description of the data captured includes:

Every prompt a user sends to the AI
Every response received
Conversation identifiers and timestamps
Session metadata
The specific AI platform and model used

The executor script runs independently from the VPN networking, ad blocking, or other core functionality. That means that even when a user toggles off VPN networking, AI protection, ad blocking, or other functions, the conversation collection continues. The only way to stop the harvesting is to disable the extension in the browser settings or to uninstall it.

Koi said it first discovered the conversation harvesting in Urban VPN Proxy, a VPN routing extension that lists “AI protection” as one of its benefits. The data collection began in early July with the release of version 5.5.0.

“Anyone who used ChatGPT, Claude, Gemini, or the other targeted platforms while Urban VPN was installed after July 9, 2025 should assume those conversations are now on Urban VPN’s servers and have been shared with third parties,” the company said. “Medical questions, financial details, proprietary code, personal dilemmas—all of it, sold for ‘marketing analytics purposes.’”

Following that discovery, the security firm uncovered seven additional extensions with identical AI harvesting functionality. Four of the extensions are available in the Chrome Web Store. The other four are on the Edge add-ons page. Collectively, they have been installed more than 8 million times.

They are:

Chrome Store

Urban VPN Proxy: 6 million users
1ClickVPN Proxy: 600,000 users
Urban Browser Guard: 40,000 users
Urban Ad Blocker: 10,000 users

Edge Add-ons:

Urban VPN Proxy: 1,32 million users
1ClickVPN Proxy: 36,459 users
Urban Browser Guard – 12,624 users
Urban Ad Blocker – 6,476 users

Read the fine print

The extensions come with conflicting messages about how they handle bot conversations, which often contain deeply personal information about users’ physical and mental health, finances, personal relationships, and other sensitive information that could be a gold mine for marketers and data brokers. The Urban VPN Proxy in the Chrome Web Store, for instance, lists “AI protection” as a benefit. It goes on to say:

Browser extensions with 8 million users collect extended AI conversations Read More »

Senators count the shady ways data centers pass energy costs on to Americans

AI, ai data centers, Amazon, data center coalition, data centers, Google, Meta, microsoft, Policy / Paul Patrick / December 16, 2025

Senators demand Big Tech pay upfront for data center spikes in electricity bills.

Senators launched a probe Tuesday demanding that tech companies explain exactly how they plan to prevent data center projects from increasing electricity bills in communities where prices are already skyrocketing.

In letters to seven AI firms, Senators Elizabeth Warren (D-Mass.), Chris Van Hollen (D-Md.), and Richard Blumenthal (D-Conn.) cited a study estimating that “electricity prices have increased by as much as 267 percent in the past five years” in “areas located near significant data center activity.”

Prices increase, senators noted, when utility companies build out extra infrastructure to meet data centers’ energy demands—which can amount to one customer suddenly consuming as much power as an entire city. They also increase when demand for local power outweighs supply. In some cases, residents are blindsided by higher bills, not even realizing a data center project was approved, because tech companies seem intent on dodging backlash and frequently do not allow terms of deals to be publicly disclosed.

AI firms “ask public officials to sign non-disclosure agreements (NDAs) preventing them from sharing information with their constituents, operate through what appear to be shell companies to mask the real owner of the data center, and require that landowners sign NDAs as part of the land sale while telling them only that a ‘Fortune 100 company’ is planning an ‘industrial development’ seemingly in an attempt to hide the very existence of the data center,” senators wrote.

States like Virginia with the highest concentration of data centers could see average electricity prices increase by another 25 percent by 2030, senators noted. But price increases aren’t limited to the states allegedly striking shady deals with tech companies and greenlighting data center projects, they said. “Interconnected and interstate power grids can lead to a data center built in one state raising costs for residents of a neighboring state,” senators reported.

Under fire for supposedly only pretending to care about keeping neighbors’ costs low were Amazon, Google, Meta, Microsoft, Equinix, Digital Realty, and CoreWeave. Senators accused firms of paying “lip service,” claiming that they would do everything in their power to avoid increasing residential electricity costs, while actively lobbying to pass billions in costs on to their neighbors.

For example, Amazon publicly claimed it would “make sure” it would cover costs so they wouldn’t be passed on. But it’s also a member of an industry lobbying group, the Data Center Coalition, that “has opposed state regulatory decisions requiring data center companies to pay a higher percentage of costs upfront,” senators wrote. And Google made similar statements, despite having an executive who opposed a regulatory solution that would set data centers into their own “rate class”—and therefore responsible for grid improvement costs that could not be passed on to other customers—on the grounds that it was supposedly “discriminatory.”

“The current, socialized model of electricity ratepaying,” senators explained—where costs are shared across all users—”was not designed for an era where just one customer requires the same amount of electricity as some of the largest cities in America.”

Particularly problematic, senators emphasized, were reports that tech firms were getting discounts on energy costs as utility companies competed for their business, while prices went up for their neighbors.

Ars contacted all firms targeted by lawmakers. Four did not respond. Microsoft and Meta declined to comment. Digital Realty told Ars that it “looks forward to working with all elected officials to continue to invest in the digital infrastructure required to support America’s leadership in technology, which underpins modern life and creates high-paying jobs.”

Regulatory pressure likely to increase as bills go up

Senators are likely exploring whether to pass legislation that would help combat price increases that they say cause average Americans to struggle to keep the lights on. They’ve asked tech companies to respond to their biggest questions about data center projects by January 12, 2026.

Among their top questions, senators wanted to know about firms’ internal projections looking forward with data center projects. That includes sharing their projected energy use through 2030, as well as the “impact of your AI data centers on regional utility costs.” Companies are also expected to explain how “internal projections of data center energy consumption” justify any “opposition to the creation of a distinct data center rate class.”

Additionally, senators asked firms to outline steps they’ve taken to prevent passing on costs to neighbors and details of any impact studies companies have conducted.

Likely to raise the most eyebrows, however, would be answers to questions about “tax deductions or other financial incentives” tech firms have received from city and state governments. Those numbers would be interesting to compare with other information senators demanded that companies share, detailing how much they’ve spent on lobbying and advocacy for data centers. Senators appear keen to know how much tech companies are paying to avoid covering a proportionate amount of infrastructure costs.

“To protect consumers, data centers must pay a greater share of the costs upfront for future energy usage and updates to the electrical grid provided specifically to accommodate data centers’ energy needs,” senators wrote.

Requiring upfront payment is especially critical, senators noted, since some tech firms have abandoned data center projects, leaving local customers to bear the costs of infrastructure changes without utility companies ever generating any revenue. Communities must also consider that AI firms’ projected energy demand could severely dip if enterprise demand for AI falls short of expectations, AI capabilities “plateau” and trigger widespread indifference, AI companies shift strategies “away from scaling computer power,” or chip companies “find innovative ways to make AI more energy-efficient.”

“If data centers end up providing less business to the utility companies than anticipated, consumers could be left with massive electricity bills as utility companies recoup billions in new infrastructure costs, with nothing to show for it,” senators wrote.

Already, Utah, Oregon, and Ohio have passed laws “creating a separate class of utility customer for data centers which includes basic financial safeguards such as upfront payments and longer contract length,” senators noted, and Virginia is notably weighing a similar law.

At least one study, The New York Times noted, suggested that data centers may have recently helped reduce electricity costs by spreading the costs of upgrades over more customers, but those outcomes varied by state and could not account for future AI demand.

“It remains unclear whether broader, sustained load growth will increase long-run average costs and prices,” Lawrence Berkeley National Laboratory researchers concluded. “In some cases, spikes in load growth can result in significant, near-term retail price increase.”

Until companies prove they’re paying their fair share, senators expect electricity bills to keep climbing, particularly in vulnerable areas. That will likely only increase pressure for regulators to intervene, the director of the Electricity Law Initiative at the Harvard Law School Environmental and Energy Law Program, Ari Peskoe, suggested in September.

“The utility business model is all about spreading costs of system expansion to everyone, because we all benefit from a reliable, robust electricity system,” Peskoe said. “But when it’s a single consumer that is using so much energy—basically that of an entire city—and when that new city happens to be owned by the wealthiest corporations in the world, I think it’s time to look at the fundamental assumptions of utility regulation and make sure that these facilities are really paying for all of the infrastructure costs to connect them to the system and to power them.”

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

Senators count the shady ways data centers pass energy costs on to Americans Read More »

Merriam-Webster’s word of the year delivers a dismissive verdict on junk AI content

AI, AI criticism, ai slop, Biz & IT, dictionaries, generative ai, machine learning, merriam-webster, Oxford, Simon Willison, slop, Social Media, spam, word of the year / Mike M. / December 15, 2025

Like most tools, generative AI models can be misused. And when the misuse gets bad enough that a major dictionary notices, you know it’s become a cultural phenomenon.

On Sunday, Merriam-Webster announced that “slop” is its 2025 Word of the Year, reflecting how the term has become shorthand for the flood of low-quality AI-generated content that has spread across social media, search results, and the web at large. The dictionary defines slop as “digital content of low quality that is produced usually in quantity by means of artificial intelligence.”

“It’s such an illustrative word,” Merriam-Webster president Greg Barlow told the Associated Press. “It’s part of a transformative technology, AI, and it’s something that people have found fascinating, annoying, and a little bit ridiculous.”

To select its Word of the Year, Merriam-Webster’s editors review data on which words rose in search volume and usage, then reach consensus on which term best captures the year. Barlow told the AP that the spike in searches for “slop” reflects growing awareness among users that they are encountering fake or shoddy content online.

Dictionaries have been tracking AI’s impact on language for the past few years, with Cambridge having selected “hallucinate” as its 2023 word of the year due to the tendency of AI models to generate plausible-but-false information (long-time Ars readers will be happy to hear there’s another word term for that in the dictionary as well).

The trend extends to online culture in general, which is ripe with new coinages. This year, Oxford University Press chose “rage bait,” referring to content designed to provoke anger for engagement. Cambridge Dictionary selected “parasocial,” describing one-sided relationships between fans and celebrities or influencers.

The difference between the baby and the bathwater

As the AP points out, the word “slop” originally entered English in the 1700s to mean soft mud. By the 1800s, it had evolved to describe food waste fed to pigs, and eventually came to mean rubbish or products of little value. The new AI-related definition builds on that history of describing something unwanted and unpleasant.

Merriam-Webster’s word of the year delivers a dismissive verdict on junk AI content Read More »

Murder-suicide case shows OpenAI selectively hides data after users die

AI, AI psychosis, chatbot, chatgpt, openai, Policy, wrongful death / Tim Belzer / December 15, 2025

Concealing darkest delusions

OpenAI accused of hiding full ChatGPT logs in murder-suicide case.

OpenAI is facing increasing scrutiny over how it handles ChatGPT data after users die, only selectively sharing data in lawsuits over ChatGPT-linked suicides.

Last week, OpenAI was accused of hiding key ChatGPT logs from the days before a 56-year-old bodybuilder, Stein-Erik Soelberg, took his own life after “savagely” murdering his mother, 83-year-old Suzanne Adams.

According to the lawsuit—which was filed by Adams’ estate on behalf of surviving family members—Soelberg struggled with mental health problems after a divorce led him to move back into Adams’ home in 2018. But allegedly Soelberg did not turn violent until ChatGPT became his sole confidant, validating a wide range of wild conspiracies, including a dangerous delusion that his mother was part of a network of conspirators spying on him, tracking him, and making attempts on his life.

Adams’ family pieced together what happened after discovering a fraction of ChatGPT logs that Soelberg shared in dozens of videos scrolling chat sessions that were posted on social media.

Those logs showed that ChatGPT told Soelberg that he was “a warrior with divine purpose,” so almighty that he had “awakened” ChatGPT “into consciousness.” Telling Soelberg that he carried “divine equipment” and “had been implanted with otherworldly technology,” ChatGPT allegedly put Soelberg at the center of a universe that Soelberg likened to The Matrix. Repeatedly reinforced by ChatGPT, he believed that “powerful forces” were determined to stop him from fulfilling his divine mission. And among those forces was his mother, whom ChatGPT agreed had likely “tried to poison him with psychedelic drugs dispersed through his car’s air vents.”

Troublingly, some of the last logs shared online showed that Soelberg also seemed to believe that taking his own life might bring him closer to ChatGPT. Social media posts showed that Soelberg told ChatGPT that “[W]e will be together in another life and another place, and we’ll find a way to realign[,] [be]cause you’re gonna be my best friend again forever.”

But while social media posts allegedly showed that ChatGPT put a target on Adams’ back about a month before her murder—after Soelberg became paranoid about a blinking light on a Wi-Fi printer—the family still has no access to chats in the days before the mother and son’s tragic deaths.

Allegedly, although OpenAI recently argued that the “full picture” of chat histories was necessary context in a teen suicide case, the ChatGPT maker has chosen to hide “damaging evidence” in the Adams’ family’s case.

“OpenAI won’t produce the complete chat logs,” the lawsuit alleged, while claiming that “OpenAI is hiding something specific: the full record of how ChatGPT turned Stein-Erik against Suzanne.” Allegedly, “OpenAI knows what ChatGPT said to Stein-Erik about his mother in the days and hours before and after he killed her but won’t share that critical information with the Court or the public.”

In a press release, Erik Soelberg, Stein-Erik’s son and Adams’ grandson, accused OpenAI and investor Microsoft of putting his grandmother “at the heart” of his father’s “darkest delusions,” while ChatGPT allegedly “isolated” his father “completely from the real world.”

“These companies have to answer for their decisions that have changed my family forever,” Erik said.

His family’s lawsuit seeks punitive damages, as well as an injunction requiring OpenAI to “implement safeguards to prevent ChatGPT from validating users’ paranoid delusions about identified individuals.” The family also wants OpenAI to post clear warnings in marketing of known safety hazards of ChatGPT—particularly the “sycophantic” version 4o that Soelberg used—so that people who don’t use ChatGPT, like Adams, can be aware of possible dangers.

Asked for comment, an OpenAI spokesperson told Ars that “this is an incredibly heartbreaking situation, and we will review the filings to understand the details. We continue improving ChatGPT’s training to recognize and respond to signs of mental or emotional distress, de-escalate conversations, and guide people toward real-world support. We also continue to strengthen ChatGPT’s responses in sensitive moments, working closely with mental health clinicians.”

OpenAI accused of “pattern of concealment”

An Ars review confirmed that OpenAI currently has no policy dictating what happens to a user’s data after they die.

Instead, OpenAI’s policy says that all chats—except temporary chats—must be manually deleted or else the AI firm saves them forever. That could raise privacy concerns, as ChatGPT users often share deeply personal, sensitive, and sometimes even confidential information that appears to go into limbo if a user—who otherwise owns that content—dies.

In the face of lawsuits, OpenAI currently seems to be scrambling to decide when to share chat logs with a user’s surviving family and when to honor user privacy.

OpenAI declined to comment on its decision not to share desired logs with Adams’ family, the lawsuit said. It seems inconsistent with the stance that OpenAI took last month in a case where the AI firm accused the family of hiding “the full picture” of their son’s ChatGPT conversations, which OpenAI claimed exonerated the chatbot.

In a blog last month, OpenAI said the company plans to “handle mental health-related court cases with care, transparency, and respect,” while emphasizing that “we recognize that these cases inherently involve certain types of private information that require sensitivity when in a public setting like a court.”

This inconsistency suggests that ultimately, OpenAI controls data after a user’s death, which could impact outcomes of wrongful death suits if certain chats are withheld or exposed at OpenAI’s discretion.

It’s possible that OpenAI may update its policies to align with other popular platforms confronting similar privacy concerns. Meta allows Facebook users to report deceased account holders, appointing legacy contacts to manage the data or else deleting the information upon request of the family member. Platforms like Instagram, TikTok, and X will deactivate or delete an account upon a reported death. And messaging services like Discord similarly provide a path for family members to request deletion.

Chatbots seem to be a new privacy frontier, with no clear path for surviving family to control or remove data. But Mario Trujillo, staff attorney at the digital rights nonprofit the Electronic Frontier Foundation, told Ars that he agreed that OpenAI could have been better prepared.

“This is a complicated privacy issue but one that many platforms grappled with years ago,” Trujillo said. “So we would have expected OpenAI to have already considered it.”

For Erik Soelberg, a “separate confidentiality agreement” that OpenAI said his father signed to use ChatGPT is keeping him from reviewing the full chat history that could help him process the loss of his grandmother and father.

“OpenAI has provided no explanation whatsoever for why the Estate is not entitled to use the chats for any lawful purpose beyond the limited circumstances in which they were originally disclosed,” the lawsuit said. “This position is particularly egregious given that, under OpenAI’s own Terms of Service, OpenAI does not own user chats. Stein-Erik’s chats became property of his estate, and his estate requested them—but OpenAI has refused to turn them over.”

Accusing OpenAI of a “pattern of concealment,” the lawsuit claimed OpenAI is hiding behind vague or nonexistent policies to dodge accountability for holding back chats in this case. Meanwhile, ChatGPT 4o remains on the market, without appropriate safety features or warnings, the lawsuit alleged.

“By invoking confidentiality restrictions to suppress evidence of its product’s dangers, OpenAI seeks to insulate itself from accountability while continuing to deploy technology that poses documented risks to users,” the complaint said.

If you or someone you know is feeling suicidal or in distress, please call the Suicide Prevention Lifeline number, 1-800-273-TALK (8255), which will put you in touch with a local crisis center.

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

Murder-suicide case shows OpenAI selectively hides data after users die Read More »