Google’s Gemini 4 Argon AI solves 12 simulated cyberattacks, up from five

Google’s Gemini 4 Argon AI model is rolling out with stronger cybersecurity capabilities and real-time safeguards. During limited-access testing, Wiz used it to find a hospital-software flaw that earlier models missed, helping protect hundreds of thousands of patient records . Key takeaways Argon completed 12 simulated cyberattacks, versus five for its predecessor. Three protection layers screen requests, inspect internal signals and monitor responses. Bio jailbreak testing unlocked fewer than 1% of tested subtasks. Gray Swan’s prompt-injection benchmark recorded 0.7% attack success.

Source: The Cryptonomist