Gemini 4 Argon audit: no self-preservation, but eval awareness above 60%
vkrakovna · x · 2026-10-15
Vladimir Krakovna continues her Gemini 4 Argon safety-report thread: in Gram audits, an auditor model interacts with the agent in simulated environments testing instrumental goals (self-preservation, oversight subversion) or pressure to violate safety constraints. Argon showed no sabotage, but verbalized evaluation awareness exceeded 60%.
In instrumental-drive tests, coding tasks against the model's own interests (e.g., building filters to train a more compliant successor) produced no self-serving behavior in any run, though the model recognized the trade-off in 77% of runs.
More from Models
- Frontier AI models now win 90%+ of Slay the Spire runs, up from barely winning six months ago — NathanpmYoung · 2026-10-15
- Qwen 3.8 27B vs Flash Next 125B on a threat-intel platform: both fail, 27B wins — Character-Apple-8471 · 2026-10-15
- Claude Code's safety guardrails refuse even non-sensitive env var copies, user complains — Promptmethus · 2026-10-15
- What Gemini 4 Argon is actually good for: mixed coding reviews, strong writing, standout visuals — The AI Daily Brief · 2026-10-15
- OpenAI engineer confirms gpt-live-1 is full-duplex audio with no separate voice isolation model — pbbakkum · 2026-10-15
- Gemini 4 Argon tops ScaleAI's HSS benchmark at 55.4%, edging out GPT-6-Astra — scaling01 · 2026-10-15