Anthropic had an unusual few days. A content management system misconfiguration left roughly 3,000 internal assets publicly accessible, including draft blog posts describing a model the company had not yet announced. That model is called Claude Mythos, internally codenamed Capybara. It sits above the existing Opus tier and, by Anthropic’s own description in the leaked drafts, represents a “step change” in capability.

The company confirmed the model exists and blamed the exposure on human error in its CMS configuration. It has since secured the data.

What the Leak Revealed

The leaked documents describe Claude Mythos as “by far the most powerful AI model we’ve ever developed,” with substantially higher benchmark scores than Claude Opus 4.6 across software coding, academic reasoning, and cybersecurity tasks.

That last category is where it gets complicated. Anthropic’s own internal drafts warn the model could significantly heighten cybersecurity risks. Specifically, the documents describe the model as capable of “autonomous vulnerability hunting” across complex, interconnected codebases, and note that unlike current models that require human guidance to verify bugs, Mythos can reportedly find, verify, and execute targeted exploits without human intervention.

The leaked text states the model is “currently far ahead of any other AI model in cyber capabilities” and warns it “presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.”

That language hit markets hard. Cybersecurity stocks sold off sharply after the leak, with CrowdStrike falling 7%, Palo Alto Networks down 6%, and Zscaler losing 4.5%. The Global X Cybersecurity ETF dropped 4.5%, bringing its year-to-date decline to over 21%.

A New Model Tier Is Coming

Claude Mythos is not yet available. Anthropic confirmed it is in limited early access and still too expensive for general release. The leaked materials suggest it will eventually sit above Opus in the model lineup, though pricing and availability details were not disclosed.

For businesses already working with Claude across APIs, the enterprise tier, or tools like Claude Code, this signals that a meaningfully more capable model is in the pipeline. Whether that excites you or concerns you probably depends on how you’re planning to deploy it.

What This Means for Business

The capability curve is steepening. The gap between what AI can do now and what was possible 12 months ago is already significant. Claude Mythos, if the leaked benchmarks are accurate, represents another major jump. Businesses that are still sitting on the sidelines of AI adoption are not just falling behind current capabilities, they are falling behind the next wave too.

Cybersecurity teams need to update their threat models. The concerns raised in the leaked Anthropic documents are real. Autonomous vulnerability discovery changes the economics of cyberattacks. Tools that previously required nation-state-level sophistication could become accessible to a much wider range of adversaries. If your security posture assumes human-paced attacks, that assumption needs revisiting.

Responsible deployment matters more at the frontier. The market reaction is partly rational fear and partly panic, but it underscores something worth taking seriously: the most capable AI systems require genuine governance frameworks, not just a use policy document. Businesses deploying AI agents in sensitive workflows should be thinking now about access controls, audit logs, and containment boundaries, not after the next generation of models ships.

Enterprise AI is moving fast regardless of how you feel about it. The more important takeaway for most business leaders is not the security risk specifically. It’s the reminder that the pace of progress in frontier AI is not slowing. Mythos is a preview of where things are headed. Companies that have built internal capability to understand and deploy AI, even at today’s model capability levels, will be far better positioned to adopt what comes next.

If this is the kind of problem agents can help with, the free Working With Claude field guide is the practical next step. Thirty-two pages, no fluff. Get the free guide.

The Mythos leak is a reminder that the frontier is moving fast. Whether that’s an opportunity or a threat mostly depends on how prepared your organization is to engage with it.