Is Anthropic Restricting Mythos to Protect the Internet or Itself?

This week, Anthropic announced it restricted the launch of its newest model, called Mythos, citing its exceptional ability to uncover security flaws in software used globally.

Rather than releasing Mythos to the public, the frontier lab plans to share it with a select group of large companies and organizations that manage critical online infrastructure, from Amazon Web Services to JPMorgan Chase. OpenAI is reportedly considering a similar approach for its next cybersecurity tool. The stated rationale is to let these major enterprises stay ahead of malicious actors who might use advanced LLMs to break into secure software.

But the word "enterprise" in the sentence above hints that this release strategy may involve more than just cybersecurity—or hype around model capabilities.

Dan Lahav, CEO of the AI cybersecurity lab Irregular, told TechCrunch in March, before Mythos was unveiled, that while AI tools discovering vulnerabilities matters, the real value of any weakness to an attacker depends on many factors, including how they can be combined.

"The question I always have in my mind," Lahav said, "is whether they found something that is exploitable in a highly meaningful way—either on its own or as part of a chain."

Anthropic says Mythos can exploit vulnerabilities far more effectively than its previous model, Opus. But it remains unclear whether Mythos truly represents the ultimate cybersecurity model. Aisle, an AI cybersecurity startup, claims it replicated much of what Anthropic says Mythos achieved using smaller, open-weight models. Aisle's team argues these results show there is no single deep-learning model for cybersecurity; rather, effectiveness depends on the specific task.

Given that Opus was already considered a game-changer for cybersecurity, frontier labs may have another reason to limit releases to big organizations: it creates a flywheel for large enterprise contracts while making it harder for competitors to copy their models through distillation—a technique that uses frontier models to train new LLMs at low cost.

"This is marketing cover for the fact that top-tier models are now gated by enterprise agreements and no longer available for small labs to distill," David Crawshaw, a software engineer and CEO of the startup exe.dev, suggested in a social media post. "By the time you and I can use Mythos, there will be a new top-tier version that is enterprise-only. That treadmill helps keep enterprise dollars flowing (which is most of the dollars) by relegating distillation companies to second rank," Crawshaw said.

That analysis aligns with what we're seeing in the AI ecosystem: a race between frontier labs building the largest, most capable models, and companies like Aisle that rely on multiple models and see open-source LLMs—often from China and often allegedly developed through distillation—as a path to economic advantage.

The frontier labs have taken a harder stance on distillation this year. Anthropic has publicly revealed what it calls attempts by Chinese firms to copy its models, and three leading labs—Anthropic, Google, and OpenAI—have teamed up to identify distillers and block them, according to a Bloomberg report.

Distillation threatens the business model of frontier labs because it eliminates the advantages gained by deploying massive capital to scale. Blocking distillation is already a worthwhile effort, but the selective release approach also gives labs a way to differentiate their enterprise offerings as this category becomes key to profitable deployment.

Whether Mythos or any new model truly endangers internet security remains to be seen, and a cautious rollout of the technology is a responsible way forward.

Anthropic did not respond to our questions about whether the decision is also tied to distillation concerns at press time, but the company may have found a clever way to protect the internet—and its bottom line.