CrowdStrike Tests Claude Mythos for Vulnerability Detection

Agentic AI , Artificial Intelligence & Machine Learning , Governance & Risk Management CrowdStrike Tests Claude Mythos for Vulnerability Detection Early Tests of New Anthropic AI Model Show Fast Detection, Better Flaw Correlation Michael Novinson (MichaelNovinson) • April 10, 2026     Share Post Share Credit Eligible Get Permission Daniel Bernard, chief business officer, CrowdStrike (Image: CrowdStrike) Claude Mythos Preview launch partner CrowdStrike found the cutting-edge artificial intelligence model dramatically reduced the time required to detect vulnerabilities, said Chief Business Officer Daniel Bernard. See Also: AI Security Risks Rise With Agentic Systems Early testing by CrowdStrike has shown Mythos Preview improves both the speed and the contextual linkage of vulnerability discovery, Bernard said, meaning defenders can not only identify flaws faster but also understand how they connect across systems. CrowdStrike and Palo Alto Networks are the only pure-play cybersecurity vendors included in Anthropic's Project Glasswing initiative for Mythos Preview (see: Claude Mythos Preview Creates Early Edge for Cyber Titans). "The model so far is really best at shrinking the amount of time it takes to detect vulnerabilities," Bernard told ISMG. "That's I think the most noteworthy finding so far." Quantifying Claude Mythos Preview Through Performance Metrics Security responsibility for AI is shared across the ecosystem rather than limited to model developers, and Bernard said independent software vendors such as CrowdStrike must play a central role in ensuring customers can adopt AI safely. This reflects a broader industry shift in which security is becoming a foundational enabler of AI adoption rather than a secondary consideration, Bernard said. "That responsibility for security is not only at the model level but is also the responsibility of ISVs like CrowdStrike to help customers adopt the technology securely and also to give defenders the advantage over adversaries," Bernard said. CrowdStrike is currently testing Mythos Preview's capabilities in controlled internal environments focusing on understanding both benefits and risks, and Bernard said a key priority over the coming months is to quantify the model’s impact through performance metrics. This includes evaluating how it enhances their platform and assessing potential misuse by adversaries, Bernard said. "The power of the model is something that Anthropic have been talking to us about for some time," Bernard said. "We have a partnership with them. And I think this was the moment that both of us wanted the industry to know that the more AI you deploy, the more security you need." As AI reshapes vulnerability management and secure development, Bernard said organizations will need new guidelines for maintaining secure and hygienic environments, as well as updated approaches to both code production and legacy system protection. While AI clearly accelerates analysis and decision-making, Bernard said human expertise is still required to interpret results and implement controls. "I think number one, it helps in identifying vulnerabilities that were either hard to identify before or require expertise and immense amounts of time to uncover," Bernard said. Compressing the Time Between Discovery, Remediation The entire operational tempo of cybersecurity is increasing, which Bernard said will require new frameworks, tools and approaches. Traditional metrics such as mean time to detect and respond will need to be redefined in an AI-driven environment, he said. As AI accelerates detection and response, new benchmarks will likely emerge to reflect the increased speed and complexity of modern threats. "I think we're entering a new chapter of security," Bernard said. "One that is an AI-accelerated chapter." Organizations have traditionally struggled with delays between vulnerability discovery and remediation, but Bernard said AI could compress this timeline significantly, potentially enabling near real-time patching. Even in cases where patching is not feasible, Bernard said AI-driven visibility can support compensating controls to mitigate risk (see: Why Claude Mythos Shifts Focus From Finding to Fixing Bugs). "You need to have the right defense, you need to have the right weapons to counter the attack of adversaries," Bernard said. If attackers can act almost immediately after a vulnerability is identified, traditional defense strategies may become insufficient. Bernard emphasized that defenders must adopt equally advanced tools and approaches to keep pace. This includes leveraging AI not just for detection but for proactive defense and rapid response, Bernard said. "Our customers will be at the forefront of protection, something that we stand for and we've delivered since the beginning of the company," Bernard said.