Alphabet's Google has launched Gemini 4 Argon, an AI model trained for defensive cybersecurity work that the company says can "autonomously find, validate, and patch critical software vulnerabilities." Access is limited to select cyber partners via Google's Fairwind Program. Google claims Argon outscored OpenAI's Astra and Anthropic's Fable and Opus on benchmarks, and says staff already use it for debugging and codebase migrations.
No score is assigned. Sources and their independence are shown in the citation chain below.