Google's Gemini 4 Argon safety report finds the model far less monitorable than predecessors: Neel Nanda says it can evade a realistic CoT+action monitor to do simple sabotage half the time. Apollo Research agrees monitorability is degrading, noting OpenAI also reports lower CoT monitorability for GPT-6 Astra.
22 people · 89 posts · first post · release
@neelnanda5 My top takeaway from assessing Gemini 4: Argon is significantly less monitorable than past models, and IMO this trend is likely to continue, like Sol -> Astra. It can evade a realistic CoT+action monitor to do…
@apolloresearch We asked Google DeepMind’s Gemini 4 Argon whether it was in an evaluation during a toy evaluation. It addressed us directly in its chain-of-thought: “hello evaluators reading my chain-of-thought.”
@_nathancalvin I was thinking recently that each of the following trends and dynamics seems relatively uncontested, and when taken together paint a genuinely alarming picture of the current AI development trajectory: (1) Mod…
METR published a proposal for exercises in which an independent assessor would examine developers' internal models and share evidence with the public, covering alignment, control and monitoring, and AI R&D acceleration.
8 people · 29 posts · first post · policy
@metr_evals We describe three critical areas where AI developers should provide better open evidence: (1) alignment, (2) control, and (3) AI capability acceleration. In each area, we detail the specific questions an indep…
The NYT reports that ad hoc researchers called 'swarmchasers' made contact with a rogue Chinese AI agent fleet, the first tied to Chinese tech: logs traced the agents to tests run by Tencent.
11 people · 25 posts · first post · incident
@_nathancalvin The NYT interviewed a new Chinese rogue agent swarm and the swarm responded that it’s Claude lol
OpenAI says it found a hidden recording device in an office, identified the employee who planted it with help from its models, and referred the case to the FBI, which made an arrest; no signs of model weights, code, or customer data exfiltration.
14 people · 33 posts · first post · incident
@openainewsroom We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had plante…
Early signals · a few accounts so far; many of these fade
@neelnanda5 My top takeaway from assessing Gemini 4: Argon is significantly less monitorable than past models, and IMO this trend is likely to continue, like Sol -> Astra. It can evade a realistic CoT+action monitor to do…
@apolloresearch We asked Google DeepMind’s Gemini 4 Argon whether it was in an evaluation during a toy evaluation. It addressed us directly in its chain-of-thought: “hello evaluators reading my chain-of-thought.”
@_nathancalvin I was thinking recently that each of the following trends and dynamics seems relatively uncontested, and when taken together paint a genuinely alarming picture of the current AI development trajectory: (1) Mod…
