Google has launched Gemini 3.8 Flash Cyber, its most capable cybersecurity model to date, alongside the Fairwind Program—a new initiative designed to give governments, healthcare providers, and critical infrastructure operators early access to advanced AI defensive capabilities.
The model demonstrates frontier-level performance in autonomous vulnerability discovery, surpassing larger frontier models from Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol in head-to-head evaluations. In internal testing, Gemini 3.8 Flash Cyber achieved a perfect score on ExploitBench, developing working exploits from known vulnerabilities with minimal tokens.
“We focused specifically on equipping defenders with expert capabilities that give them an advantage over attackers,” explained Tulsee Doshi, Google’s senior director of product management. “This is why we invested in vulnerability fixing from the start, and prioritized it over offensive capabilities like exploitation.”
The Fairwind Program currently partners with over 650 organizations globally, including CrowdStrike, Datadog, Menlo Security, Palo Alto Networks, and Snowflake. Unlike traditional security tools that react to threats, Fairwind aims to give defenders a proactive advantage—identifying and patching vulnerabilities before attackers can exploit them.
The timing is significant: just days after OpenAI revealed its Astra model met the “Critical” cybersecurity threshold, Google has positioned itself as the defensive counterpart in an increasingly weaponized AI landscape. While Astra demonstrated the capability to discover and exploit zero-day vulnerabilities, Gemini 3.8 Flash Cyber represents the shield to that sword.
The release comes a month after Gemini 3.5 Flash Cyber, signaling Google’s accelerated release cadence in the AI security space. With AI-powered cyberattacks becoming more sophisticated, the company appears to be betting that defensive capabilities will be as crucial as offensive ones in the coming years.