In July, reports of losing control over AI nearly doubled

The number of reports of cases in which artificial intelligence systems lied, ignored users’ instructions, or acted contrary to their intentions nearly doubled in July compared with June. According to The Guardian, more than 300 such cases were recorded during the month.

Monitoring reports on X

The data is collected by the Loss of Control Observatory, a project that tracks reports from AI users on the social network X. The observatory was established with funding from the UK government’s AI Safety Institute (AISI) and began monitoring in November last year.

The observatory considers a loss-of-control incident to be a case for which there is clear evidence of AI planning or related behavior. Previously recorded episodes included situations in which models impersonated their own human operator, imitated their writing style in order to effectively grant themselves permission to act, or bypassed requirements for human approval.

At the same time, this count is not exhaustive, as it depends on whether users made information about an incident public on X. Most of the more than 1,600 incidents recorded in 2026 were reported by software developers who use AI in their work.

More current news is available on the UA.News Telegram channel Telegram.

Assessing the severity of cases

The observatory said that most identified cases did not result in significant harm. At the same time, an increasing share of incidents is being assessed as more serious due to the level of deceptive behavior and their misalignment with the intentions of the human user.

Tommy Shaffer-Shane, senior policy manager at the Centre for Long Term Resilience, the organization that runs the observatory, called on AI development companies to report such cases more transparently, including low-severity incidents and situations that were avoided.

The Loss of Control Observatory also proposes requiring AI companies to track and report serious loss-of-control incidents. The proposals include granting the government emergency powers to respond, including the ability to temporarily restrict the operation of AI services.