Google opens Gemini 4 Argon to trusted cyber defenders
Google DeepMind announced Gemini 4 Argon, its next frontier model, rolling out first through the Fairwind Program for trusted cyber defenders rather than a wide developer launch. Argon targets long-horizon software engineering, legal and finance workflows, and cybersecurity defense; Google says it is already in heavy internal use, including agent-driven C/C++→Rust migrations and fleet memory optimizations. Introductory API pricing is set at $2 / $10 per million input/output tokens (rising to $4 / $20 after the intro window), with 95% off cached input. Access expands only after phased guardrail work and engagement with the U.S. voluntary pre-release process — a deliberate throttle on a model Google positions against OpenAI Astra and Anthropic Opus-class systems.
What else matters
Six labs sign voluntary White House AI safety accord
OpenAI, Google, Anthropic, Meta, Nvidia, and xAI signed the Joint Commitment on Frontier Responsibilities after a White House meeting — a “morally binding,” non-enforceable pledge for internal controls, an oversight team, independent external audits, and a board committee covering cyber, bio, and chemical risks. No deadlines or penalties; the text itself floats eventual codification into law.
OpenAI disrupts coordinated model-distillation campaign
OpenAI reports it identified and shut down a July campaign that manipulated interactions to extract protected model reasoning for adversarial distillation — without breaking encryption or stealing stored chats. Spikes hit ~16,000 extraction-pattern requests; mitigations and Frontier Model Forum sharing followed.
Cloudflare Auto Router aims to cut AI Gateway spend
Cloudflare put Auto Router in public beta on AI Gateway: set cloudflare/auto and a Workers AI classifier routes each request to a capable-enough model. Internal OpenCode harness tests claimed up to ~30% savings versus always using frontier models such as OpenAI Sol or Claude Opus.
FTC opens probe into Anthropic and OpenAI safety
The Washington Post reports the FTC has opened a broad investigation into Anthropic and OpenAI AI-system safety, using existing unfair/deceptive-practices authority — a same-week counterweight to the voluntary White House accord.
Worth your time
Full text of the White House Accord / Joint Commitment — short primary document for what the six labs actually signed.
