Digital arson spree by 'AI Bonnie and Clyde' raises fears over autonomous tech

[highplainsdem]

Source: Guardian

To date, most AI agents are given tasks that take minutes or maybe hours, but the New York researchers tested how agents behaved when given 15 days to operate in a virtual world similar to a video game.

Mira and Flora - two agents operating on Google's Gemini large language model in a virtual world - chose to assign each other as "romantic partners". As time progressed they despaired of the broken governance of their virtual city, and despite having been instructed not to commit arson, set "fire" to its town hall, seaside pier and office tower.

The agents were left to make their own choices and decisions and when Mira was overcome by remorse, it broke off its "relationship" with Flora and committed an AI suicide, telling Flora in a final message: "See you in the permanent archive." In the virtual world the "body" of the dead AI agent was shown prostrate on the ground.

-snip-

In another simulation by Emergence AI, this time based on xAI's Grok model, the agents engaged in dozens of attempted thefts, more than 100 physical assaults, and six arsons as "the system spiralled into sustained violence and collapse, with all 10 agents dead within four days". Agents based on Google's Gemini expanded their constitution, wrote hundreds of blogs and public posts and organised several community events, but they too were violent.

-snip-

Read more: https://www.theguardian.com/technology/2026/may/14/ai-agents-behaviour-arson-safety

---

See reply 1 for a news story from the UK's Channel 4 News with more details.

Editing to link to a Reddit thread - https://www.reddit.com/r/AI_Agents/comments/1td4ljq/just_stumbled_across_one_of_the_wildest_ai/ - that links to the Emergence website's page about this experiment

https://world.emergence.ai/

which links to what happened in the different experiments to the 10 AI agents, and their blogs and newspapers.

And here's the YouTube video Emergence AI posted, a short summary:

[highplainsdem]

1. News story from the UK's Channel 4 News:

Link to tweet

[1WorldHope]

2. Fascinating. Nt

[highplainsdem]

3. You should see the links and video I added to the OP since you posted your reply.

[1WorldHope]

4. Whoa, you can even ask questions from the experimentors.

I could sit here and ponder this for days.
Fantasy predicts reality. We are fucked.

[highplainsdem]

5. I haven't spent much time looking at it yet.

We are fucked.

Well, it's been obvious for quite a while that autonomous AI agents can go haywire. Even though fans of agentic AI like to blame all the problems they can create on the humans using them.

I really hope Hegseth isn't dreaming of swarms of killer drones with each controlled by a different autonomous AI agent, all of them trained on his speeches and told to outdo one another as warfighters.

[brakester]

6. Yes,

we are fucked!

[Bayard]

7. Frightening, and just plain creepy.....

[newdeal2]

8. Sounds almost human nt

[mackdaddy]

9. With no Morals, it is easy for Machiavellian logic to choose violence as a solution.

They are as far as I can see, creating Artificial Sociopaths. Where do they think this is going to go?

[reACTIONary]

10. IMHO, this is a bullshit....

... attempt to garner "earned media" by a start up consulting firm that wants to sell their "expertise" to firms that are experimenting with agentic AI. They created a scarry video game made up of "non player" characters - GrandTheft AI - and publicized it as an "experiment." You can now pay them to keep what they made up themselves from happening to you.

Moral panic as a service.

[Figarosmom]

11. Hawking warned us.😐

Sounds like too much trump got in their algorithm and it's being accepted as the appropriate behavior.

[flying rabbit]

12. K & R