(Credit: Samuel Boivin/NurPhoto via Getty Images)
Reports have come to light over the past year about how cybercriminals, including state-sponsored actors, have been able to use frontier AI models from AI firm Anthropic to find exploitable vulnerabilities within code, even in the mature and well-maintained systems of large corporations. Now, the OpenAI rival says it is rolling out similar vulnerability-identifying AI capabilities for its customers.
Anthropic is launching Claude Code Security, a new cybersecurity capability for Claude Code, its popular AI coding tool. The company says the tool can scan codebases for security vulnerabilities and suggest targeted software fixes for human review.
The company claims the tool differs from traditional security software because it does not rely on rule-based pattern matching—where code is analyzed and compared to known vulnerabilities—but instead “reasons through your code like a security researcher.”
Issues identified by the tool will be assigned severity ratings to help security teams prioritize, as well as “confidence rankings” indicating how certain the system is in its assessment of each risk, after automatically reexamining findings for false positives. The tool will not make changes to code directly but will add identified issues to a dashboard for human security teams to assess and act on.
Claude Code Security is now available in limited research preview for Enterprise and Team customers. Those maintaining open-source code repositories can also apply for free, expedited access.
Anthropic is not alone in rolling out this type of tool. OpenAI began beta testing Aardvark, an agentic security researcher powered by GPT-5, in October.
Some may be betting traditional cybersecurity companies could be negatively impacted by the rise of tools like Anthropic’s new offering. SiliconANGLE reports that cybersecurity firm CrowdStrike Holdings closed the trading session down almost 8% after Claude Code Security was announced, while Cloudflare fell just over 8%.
In a November interview with PCMag, Boris Cherny, Head of Claude Code, hinted that future models "are gonna run for a longer period of time without human intervention" and teased increased integrations with other AI models.