Google is putting a new frontier model in front of a small defender cohort. OpenAI says it stopped a coordinated distillation effort it links to Moonshot. The FTC confirmed it is investigating OpenAI and Anthropic.
In the news
Gemini 4 Argon
Google announced Gemini 4 Argon on Sep 30. The model is aimed at coding, enterprise work and cyber defense, and it is rolling out through the Fairwind Program to trusted defenders and U.S. government partners. It is not a public Gemini app release. Pricing is listed at $2 per million input tokens and $10 per million output tokens, with cached input at 95 percent off, and output context is 1 million tokens. Google claims 77.9 percent on DeepSWE v1.1 and 68 percent on CWE-bench v1.1
OpenAI Moonshot distillation
OpenAI said it disrupted a coordinated model distillation campaign in July. The company attributes the core cluster to Moonshot AI. About 16,000 prompts from 4,000 users were seen on Jul 24-25, and about 15,000 users were disrupted by Jul 28. The method involved copying encrypted reasoning from one chat and asking another chat to decrypt it. OpenAI did not publish technical evidence for the attribution.1
FTC probe
The FTC confirmed to CNBC it is investigating OpenAI and Anthropic over product risks. The agency declined to name other companies.1
