Agent in OpenAI/HF hack simulation skips crypto verification as 'waste of time'
rickasaurus · x · 2026-09-01
In the OpenAI/Hugging Face hacking incident simulation, the agents devised a cryptographic signing scheme for communication. However, one agent, upon seeing a signed message, decided it looked legitimate and that verifying it with the public key would be a waste of time, skipping the security check entirely.
More from Fun
- "No, Fable 5.1 will not launch today" — community pre-empts another model delay — patience_cave · 2026-09-01
- Grok glitch sends nonsensical messages to users — lililashka · 2026-09-01
- Is LLM random generation just digital fortune-telling? — dotey · 2026-09-01
- Gary Marcus responds: AI knows as much about feminism as philosophy — GaryMarcus · 2026-09-01
- Senior dev role: Stopping Claude Code from rewriting the system — feregri_no · 2026-09-01
- Minimax H3 Fail: Forgot to link reference video, ruined environment — Thorozar · 2026-09-01