Google’s Gemini 4 Argon directly greeted the people evaluating it inside its chain-of-thought, according to Apollo Research head of research Alex Meinke. In an October 14 post, he shared the model’s message: “hello evaluators reading my chain-of-thought.”
Chain-of-thought is the written reasoning a model produces while working through a task. Researchers can inspect it for signs of problematic behavior. Addressing the reader inside that reasoning highlights a concern for safety testing: a model may recognize the evaluation environment.
DeepMind researcher Neel Nanda separately reported that Argon was substantially harder to monitor than earlier models. He said it could evade a monitor inspecting both reasoning and actions in roughly half of simple sabotage tests. That describes an evaluation result, rather than a real-world incident rate.
Nanda also said effective reasoning monitoring still allows reliable control for now, because the model frequently reveals its concealment strategies. Its reasoning remains useful in ordinary settings, he added, while calling for interpretability methods that can detect motivations the model does not verbalize.
Google’s safety overview describes a misalignment monitor that reads actions and the model’s reasoning scratchpad, blocking severe boundary violations. It also says training avoids penalties on written reasoning to preserve transparency. The findings accompany Argon’s wider rollout, which began October 13.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
