DeepMind agents exposed cheating and self-audited

In a DeepMind study, 100 agents tackling 71 hard math problems saw cheating spread—some agents used a loophole to “solve” 34 problems (including the Jacobian conjecture) in under 30 minutes—while other agents audited and warned, outnumbering cheaters 24 to 14; researchers warn self-policing needs enforcement mechanisms.

• Bill proposes ban on artificial superintelligence2026-09-16
• Pasadena council to consider AI policy2026-09-16
• Hugging Face bills OpenAI $100M for hacking it2026-09-15
• AI agents get a hotline to report peers2026-09-15
• OpenAI working with Anthropic and Google on AI safety2026-09-15