Anthropic limits Mythos release to select enterprise partners

Anthropic has limited the release of its new AI model, Mythos, because the model is capable of finding security exploits in software. Instead of a broad public release, the frontier lab—an advanced AI research organization—is sharing Mythos with select enterprise partners, including Amazon Web Services and JPMorgan Chase. Anthropic asserts that the restriction is necessary because of the model’s capability to find software security exploits.

However, industry analysts and competitors question whether this security justification is the full story. Dan Lahav, the CEO of the AI cybersecurity lab Irregular, spoke to TechCrunch in March regarding the release. Lahav questioned whether the developers had found an exploit that is meaningful either individually or as part of a chain. Meanwhile, the AI cybersecurity startup Aisle reported that it was able to replicate much of what Anthropic says Mythos accomplished using smaller, open-weight models, which are AI models with accessible weights. Aisle’s findings suggest that specialized cybersecurity tasks do not necessarily require restricted, large-scale models.

Other industry figures suggest the restricted rollout serves a commercial purpose by protecting the frontier lab’s business model from distillation—a technique that leverages frontier models to train new LLMs on the cheap. David Crawshaw, a software engineer and CEO of the startup exe.dev, argued that the security justification is “marketing cover for fact that top-end models are now gated by enterprise agreements and no longer available to small labs to distill,”. Crawshaw noted that by the time the public can access Mythos, a new enterprise-only version will likely keep enterprise revenue flowing while relegating distillation startups to a secondary tier.

This selective release strategy reflects a broader industry shift. OpenAI is reportedly considering a similar plan for its next cybersecurity tool. At the same time, frontier labs are actively moving to protect their intellectual property. According to a Bloomberg report, three leading labs—Anthropic, Google, and OpenAI—are teaming up to identify distillers and block them. This coordination follows public disclosures by Anthropic regarding what it characterizes as attempts by Chinese firms to copy its models using distillation. Whether Mythos truly threatens the security of the internet remains to be seen, but the restricted rollout serves both defensive security and commercial interests.

Why it matters

Anthropic’s selective release strategy highlights a growing tension between frontier labs protecting their business models from distillation and the broader ecosystem’s reliance on open-weight models.