Anthropic has announced the release of Claude Sonnet 4.5, an innovative AI model designed to help defenders detect, analyze, and remediate code and deployed system vulnerabilities.
Taking advantage of the growing role of AI in cybersecurity, Sonnet 4.5 was trained with a focus on code vulnerability detection and patching.
In benchmarks like Cybench and CyberGym, the model outperformed Claude Opus 4.1—its larger and more expensive predecessor—at detecting vulnerabilities, reproducing attacks, and even discovering new vulnerabilities.
Examinations showed that Sonnet 4.5 completed over 76% of challenges in multiple attempts, reducing task durations that would take human experts hours to under 40 minutes.
Anthropic noted that this advancement is an inflection point for the defensive application of AI in cybersecurity.
Outside of research benchmarks, practical partners such as HackerOne and CrowdStrike have reported more rapid, accurate security workflows with Claude.
Though the model is no substitute for experienced professionals, it augments capabilities in SOC automation, vulnerability remediation, and red-teaming.
Anthropic underscored the urgency of scaling defensive use cases so that defenders, rather than attackers, reap most from frontier AI breakthroughs.
You may also want to check out some of our other recent updates.
Wanna know what’s trending online every day? Subscribe to Vavoza Insider to access the latest business and marketing insights, news, and trends daily with unmatched speed and conciseness! 🗞️
