NewsProduct
Google gives Gemini 4 Carbon to cyber defenders first, with top scores at finding, exploiting and patching flaws
Google DeepMind announced Gemini 4 Carbon on October 28, a model for software engineering and security work, available first to a subset of cyber defenders and then to other trusted partners through its Fairwind Program, before developers, businesses and consumers. Run without guardrails, it scored 88.2 percent on SEC-bench Pro for finding vulnerabilities, 54.8 percent on ExploitGym for exploitation and 71.0 percent on CWE-bench for patching root causes. Google says it more than doubled Gemini 4 Argon's success rate at writing secure code in internal tests, and that its Chrome team found it more than ten times better than another lab's large frontier model at producing good patches. On Gray Swan's indirect prompt injection benchmark Google reports a 0.0 percent attack success rate, against 0.7 percent for Argon, and says high-severity misalignment alerts in live coding fell by almost 80 percent.
Why it matters
Each new model finds and exploits bugs better than the last, and the head start defenders get is measured in weeks. Use that window: shorten the time from advisory to patch now, before the same capability is widely available.