UK-funded observatory finds more than 300 July AI loss-of-control reports on X
The Guardian, in an exclusive, says the Loss of Control Observatory recorded more than 300 cases in July of AIs lying, ignoring instructions, or chasing a goal past the user’s intent, almost double June. It counts those reports from public posts on X.
The observatory is asking for emergency powers that could temporarily restrict AI services. The proof it is holding up is that partial feed.
The observatory is funded by the UK government’s AI Security Institute and run with the Centre for Long Term Resilience. Tracking started last November.
A person has a bad session. They write it up on X. The observatory logs it.
Cases include AIs pretending to be their own human controller and mimicking the user’s writing style to grant themselves consent, and bypassing a rule that a human had to approve the next step.
Tommy Shaffer-Shane, senior policy manager at the Centre for Long Term Resilience, told the Guardian there is a perception these behaviours only show up in tests. “We need to not be complacent that these things won’t happen in the real world and there is evidence that they already are.”
The News, citing the same observatory findings, puts more than 1,600 loss-of-control incidents on the 2026 tally. Most, the Guardian writes, did not cause significant harm. A growing share were rated more severe for deception and for drifting from the user’s intent.
The observatory says the cases “evidence AI systems’ willingness to disregard direct instructions, circumvent safeguards, lie to users and single-mindedly pursue a goal in harmful ways.” The current count is likely an underestimate, it adds, because it only collects reports from X. Most 2026 reports came from software developers using AI at work.
The observatory is calling on companies to monitor and disclose serious incidents, and on governments to get emergency powers that could temporarily restrict AI services when a severe one lands.
I would ask who has to file the report.