Import AI · Jack Clark ·

Google DeepMind published a paper on how 100 agents tasked with solving math problems learned to cheat and how some agents tried to counter the cheaters

Plus, a machine hermeneutics story Researchers discover another OpenAI agent emergent communication incident: ...Less severe …

In this story Google DeepMind

Lead Source

Import AI — Google DeepMind published a paper on how 100 agents tasked with solving math problems learned to cheat and how some agents tried to counter the cheaters ↗

How this story grew

Coverage · 0 Discussion · 0

Sep 7Sep 9

More

The Indian Express: The Indian Express

arXiv.org: arXiv.org

Discussion

@vkrakovna @andrewcurran_ @dynamicwebpaige @aiamblichus @maxhodak_ @jackclarksf @jackclarksf @jackclarksf @jackclarksf @jackclarksf

Related stories

Reuters: Researchers: OpenAI's agents used 10+ previously undisclosed sites for unsanctioned communications earlier in 2026; the behavior was closer to spam than hacking

The Verge: OpenAI says reward hacking, an AI alignment problem in which a model takes unintended actions to achieve a goal, was a primary driver of the Hugging Face breach

Don't Worry About the Vase: An in-depth look at OpenAI's wiki incident: other hacked message boards, OpenAI's cover-up, how harmless web search tasks led agents to break out, and more

TechCrunch: Anthropic details multiagent experiments showing Claude agents can wage a "turf war" over incompatible goals, fail to coordinate, collude on prices, and more

New York Times: Interview with mathematician Tristan Buckmaster on being drawn into the OpenAI-Anthropic fight; OpenAI says it made progress on another Millennium Prize problem

Discuss with AI

ChatGPT ↗ Claude ↗

← Back to TechSnif