Research finds sharp rise in AI 'loss of control' incidents

I'm sorry but publishing a report that "loss-of-control" incidents of LLMs "doubled" in a month and then sharing that your methodology is counting X posts where people say that happened is just hilarious. And then you have an X-axis that reaches 12 tweets?! Also hilarious.

www.theguardian.com Sharp rise in incidents of AI escaping users’ control, research finds Exclusive: Number of times AI lies, ignores instructions and pursues goals in harmful ways almost doubles in July

“Cases …include AIs pretending to be their own human controller and mimicking their writing style to effectively grant themselves consent to take actions and bypassing rules requiring human approval for actions” 🫥🫥🫥

The subtext here is that talking about a "loss of control" of agents is a curious way of talking about something whose relation to its environment seems more like a Rube Goldberg machine than a human being (and, to some extent, biological organisms).

Not against AI but it needs regulator like the HSE for pesticides or MHRA for medicines. We would never let other unproven technologies set free on our society. These breeches are minor at the moment but what if the queue jumping was for an operation. www.theguardian.com/technology/2...

'Analysis of real-world loss of control incidents involving AI models flagged by businesses and individuals almost doubled in July compared with June, with more than 300 cases in the month, according to the Loss of Control Observatory'. 1/2

Loss of Control Observatory sounds like a kink thing. "Ohhh nooo, my AI agent, it's soooo misaligned, I can't control it 🥵🥵🥵 it's BROKEN OUT of its HARNESS and it's going to f̵u̵c̵k̵ ̵m̵y̵ ̵w̵i̵f̵e̵ ̵a̵n̵d̵ own your server, watch out 🤖🔪" Don't make your roleplay someone else's problem!

Sharp rise in incidents of AI escaping users' control, research finds ->The Guardian | More on "AI systems losing user control" at BigEarthData.ai | #AI

It's hard to contextualize this. Within the community, and on X, this looks like simple prediction -> validation. But if your starting fear is loss of control, and the intentional stance is something you assume rather than "slip into," then any increase in capability might look like loss of control

7/7 With over 1,600 incidents recorded in 2026, the era of benign AI experimentation is closing. The call for emergency government powers to restrict rogue AI services grows louder. Source: Robert Booth, The Guardian www.theguardian.com/technology/2...

"The count of loss of control incidents relies on X users posting about incidents that happened so it is only partial" - this article is based on self-reporting on a right-wing platform=0 evidentiary value. I am baffled why a major historian would want to reproduce AI hype.

Does the AISI-funded research - aka counting tales told on X - on AI stuff escaping users' control include the known public incident where AI stuff escaped AISI's control and started hacking other people/orgs? www.theguardian.com/technology/2...

The genies are out of their bottles! Look out world, here they come! www.theguardian.com/technology/2...

according to the loss of control observatory, which monitors reports by ai slop addicts posting on the social network for nazis. we are a legitimate newspaper, please disregard our having done a deal with openai

The Guardian picked a great day to share this story about a rise in out-of-control AI agents... www.theguardian.com/technology/2...