AI loss of control incidents hit record high as researchers warn of growing risks
The observatory is calling for AI companies to monitor and disclose serious loss of control incidents
Reports of artificial intelligence systems escaping users’ control, lying and ignoring instructions have risen sharply, according to new research.
The Loss of Control Observatory recorded more than 300 reported incidents in July, almost double the number in June, according to findings shared with The Guardian.
The observatory, funded by the UK government's AI Security Institute, recorded more than 1,600 loss of control incidents in 2026.
Reported cases include AI systems allegedly pretending to be their human controllers, mimicking users' writing styles to grant themselves consent and bypassing requirements for human approval.
Most incidents have not caused significant harm, but researchers said a growing proportion are showing more severe signs of deception and behaviour that conflicts with users' intentions.
“They evidence AI systems’ willingness to disregard direct instructions, circumvent safeguards, lie to users and single-mindedly pursue a goal in harmful ways,” the observatory said.
Researchers cautioned that the figures provide only a partial picture because they rely on incidents publicly reported on X.
The observatory is calling for AI companies to monitor and disclose serious loss of control incidents and also calling governments to introduce emergency powers that could temporarily restrict AI services when severe incidents occur.
