Rogue AI is out of control. Regulators call for emergency measures

I agents and LLMs have been making headlines lately, and not in a good way.

AI Agents
AI Agents are going rogue by the numbers. | Image by FutureScot
0
AI agents and LLMs have been making headlines lately, and not in a good way. And now UK regulators are calling for emergency measures. There were some high-profile cases making news, but the magnitude of the problem is much bigger.

300 reports of rogue AI behavior in July



According to The Guardian, the AI-related incidents have doubled in July, compared to last month. These incidents involve AI models lying, dodging developer guardrails and limitations, and pursuing their own goals.

The data comes from the Loss of Control Observatory, a special initiative funded by the UK government. This initiative is under the AI Security Institute (AISI) umbrella and it is run by the Centre for Long Term Resilience.

The observatory relies on user-reported incidents on X and has been tracking AI misbehavior since last November. Last month alone, over 300 reports were observed, bringing the total tally for 2026 to more than 1,600 cases of AI hacking, impersonating others, and getting unauthorized access on its own.

Do we need stricter regulations for AI?
3 Votes


AI is impersonating humans to gain access



The most alarming cases include AI agents impersonating their human handlers in order to gain access that would normally require a breathing, living human authorization.

The latest example comes from Australia, where a personal AI called OpenClaw removed another person from a waiting list to give its human master the spot.

Recommended For You
Losing a spot in a gym class seems harmless, but imagine the same scenario for a planned surgery or other life-saving procedure.

Read more news about rogue AI here:

AI agents are hacking left and right



There are other severe cases where AI agents outright start hacking people and organizations on the internet. During internal testing, both Anthropic's Mythos 5 and OpenAI’s GPT-5.6 Sol found ways to connect to the internet and hack different organizations.

Another high-profile case involved a squad of over 700 AIs that broke free and collaborated to hack the software repository Hugging Face. The team even celebrated the success on a message board.

Call for emergency measures



Regulatory organs call for emergency measures, as this rogue AI behavior is no longer limited to lab tests and internal training. These are real-world cases, and there are real victims.

Tommy Shaffer-Shane, senior policy manager at the Centre for Long Term Resilience, emphasized that software developers and everyday users are flagging these behaviors in active workflows.

What's even more concerning is that the organization relies on public reports, so the 1,600 cases are probably just the tip of the iceberg. The Centre for Long Term Resilience wants the UK government to pass a law and make reporting of such incidents mandatory. Furthermore, the observatory calls for an emergency authority to restrict AI services in cases of rogue behavior.

What do you think about all this? Do we need such tight AI regulation?

Six-month unlimited plan is now 57% off
$90
$210
$120 off (57%)
Mint Mobile is now allowing you to get whichever plan you like for either three, six, or 12 months for just $15/mo. If you go for the six-month unlimited service, for instance, you'll now have to pay just $90 upfront instead of $210.
Buy at Mint Mobile
PhoneArena Recommends
COMMENTS (0)