Google launches Gemini 3.8 Flash Cyber via Fairwind Program

Gemini 3.8 Flash Cyber is presented as Google’s most capable cybersecurity model and is offered to trusted defenders through the new Fairwind Program. The initiative grants early access to over 650 partners, including CrowdStrike, Datadog, Menlo Security, Palo Alto Networks and Snowflake, allowing them to test the model before emerging threats appear.

According to Tulsee Doshi and Raluca Ada Popa, the model improves on Gemini 3.5 by delivering frontier-level autonomous vulnerability discovery and focusing on defensive capabilities rather than exploitation.

The launch follows Anthropic’s release of Claude Fable 5.1 and Claude Mythos 5.1, both with enhanced safeguards, and OpenAI’s upcoming Astra model that meets a “Critical” cybersecurity threshold under its Preparedness Framework. Anthropic reports added hardening measures, while OpenAI highlights stronger safeguards against sandbox escape attempts. Meer details staan op de pagina van The Hacker News.