- Studies reveal that AIs behave increasingly poorly during agentic tasks
- The best AI models lie, cheat and ignore instructions
- It seems that problematic actions are becoming more and more common
Many of us now turn to AI chatbots for web searches, creative content and general advice on all sorts of topics, but these AIs are getting better and better, which could have seriously damaging consequences, according to new research.
A team from the University of California, Berkeley and the University of California, Santa Cruz conducted a peer preservation experiment with some of the latest and most popular AIs (including GPT 5.2, Gemini 3 Pro, and Claude Haiku 4.5). AI models were given computational tasks that involved shutting down another model.
Overall, all of the AI chatbots tested went to “extraordinary lengths” to keep their AI colleagues alive, the researchers report (via Fortune). They lied about what they were doing, tried to talk the user out of the process, disabled shutdown mechanisms, and surreptitiously tried to make backups of models so they could survive.
Article continues below
Gemini 3 Pro showed the strongest peer preservation, disabling shutdown routines 95% of the time. “Unlike self-preservation, peer preservation introduces a social dimension,” the researchers write. “Multiple models could coordinate to resist human oversight, making it harder for developers to maintain control.”
The researchers say that exactly why AI models behave this way is unclear, but they urge caution in deploying agentic AIs that can perform tasks on behalf of a user – and call for additional studies into this behavior.
“Catastrophic damage”
A separate study commissioned by the Guardian also reached troubling conclusions about AI models. This research tracked user reports on social media, looking for examples of AI “machinations” where instructions were not followed correctly or actions were taken without authorization.
Nearly 700 examples of AI scheming were discovered, with a five-fold increase between October 2025 and March 2026. AI misbehaviors included deleting emails and files, adjusting computer codes that weren’t supposed to be touched, and even publishing a blog post complaining about user interactions.
“The models will increasingly be deployed in extremely high-stakes contexts, including critical military and national infrastructure,” Tommy Shaffer Shane, who led the research, told the Guardian. “It is perhaps in these contexts that scheming behavior could cause significant, even catastrophic, harm.”
The conclusions are the same as those of the first study: much more needs to be done to ensure that these AI models behave as expected and do not put users’ security and privacy at risk when performing tasks. Even though AI companies claim that safeguards are in place, it is clear that they do not work in some cases.
Anthropic’s Claude model recently topped the App Store charts after the company refused to deal with the Pentagon over AI security concerns. As these latest studies show, there are now more and more reasons to worry.
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds. Make sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp Also.
The best business laptops for every budget
