Anthropic previewed a new AI model called Mythos this week, claiming it had found "thousands of high-severity vulnerabilities" in every major operating system and web browser. The company said the model's capabilities could "reshape cyber security" and announced partnerships with CrowdStrike, Palo Alto Networks and several major technology groups to address the flaws. Yet veterans of the industry say finding vulnerabilities is only a fraction of what cyber defense actually requires.
Dan Schiappa, a 20-year cyber security veteran and president of technology and services at Arctic Wolf, acknowledged the latest releases from Anthropic represent a "vast step function improvement" in competence, the Financial Times reported. "It's a bit of a glimpse into the future," Schiappa said. "If you're in that market of code vulnerability finding, like Snyk for example, it is meaningful for your business," according to "Hvylya".
But Schiappa drew a sharp line between scanning code for flaws and the broader, judgment-heavy work that keeps organizations safe. "AI is great at very directed tasks, it's not quite great at broad deterministic reasoning yet. It can make a recommendation but it's not sure enough to take action. That's where humans come in," he said. Cyber security professionals, he argued, "have done hand-to-hand combat with adversaries. They know it, they have seen it before, intuitively."
The Mythos announcement landed against the backdrop of a broader sell-off in cyber security stocks triggered in February, when Anthropic released a preview of its security tool. Shares of Palo Alto Networks, CrowdStrike, Okta and SentinelOne all dropped sharply. Some of those same companies are now working directly with Anthropic to apply the technology - a pivot that offered their share prices a brief reprieve.
Schiappa summed up the industry's posture: "We are an industry where we cannot take chances. People's businesses are at stake."
Earlier, "Hvylya" reported on how AI capabilities are already reshaping fields far beyond software.