Is Anthropic limiting the release of Mythos to protect the internet — or Anthropic?

Anthropic said this week that it has limited the release of its latest model, called Mythos, because it is too good at finding security holes in software that users around the world rely on.

Instead of unleashing Mythos on the public, the border laboratory will share it with a group of major companies and organizations operating critical online infrastructure, from Amazon Web Services to JPMorgan Chase.

Open AI is reportedly is considering a similar plan for its next cybersecurity tool. The ostensible idea is to give these large enterprises an advantage over bad actors who can use sophisticated LLMs to penetrate secure software.

But the “e-word” in the sentence above is a clue that this release strategy may involve more than just cybersecurity – or hype of model capabilities.

Dan Lahav, the CEO of the AI ​​cybersecurity lab Irregulartold TechCrunch in March, before the release of Mythos, that while the discovery of vulnerabilities by AI tools matters, the specific value of each weakness to an attacker depends on many factors, including how they can be used in combination.

“The question I always have in my mind,” Lahav said, “is: Have they found something that can be exploited in a very meaningful way, individually or as part of a chain?”

Anthropic says Mythos can exploit vulnerabilities much more than its previous model, Opus. But it’s not clear that Mythos is actually the all-encompassing cybersecurity model. Aisle, an AI cybersecurity startup, said it was able to replicate much of what Anthropic says Mythos achieved using smaller, open-weight models. Aisle’s team says these results show that there is no single deep learning model for cybersecurity, but instead depends on the task at hand.

Considering that Opus was already seen as a game changer in cybersecurity, there’s another reason why frontier labs might want to limit their releases to large organizations: it creates a flywheel for large enterprise contracts, while making it harder for competitors to copy their models using distillation, a technique that frontier models use to cheaply train new LLMs.

“This is marketing cover for the fact that top models are now locked into corporate agreements and are no longer available for small labs to distill,” says 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-end revenue that applies only to enterprises. That treadmill helps keep the enterprise dollars flowing (which is the bulk of dollars) by relegating distilling companies to the second tier,” Crawshaw said.

That analysis is consistent with what we’re seeing in the AI ​​ecosystem: a race between frontier labs developing the largest, most capable models, and companies like Aisle that rely on multiple models and see open source LLMs, often from China and often reportedly developed through distillation, as a path to economic advantage.

Border labs have taken a tougher stance on distilling this year, with Anthropic publicly revealing what it says is Chinese companies trying to copy its models, and three leading labs – Anthropic, Google and OpenAI – working together to identify and block distillers, according to a Bloomberg report.

Distillation threatens the business model of frontier laboratories because it eliminates the benefits associated with using enormous amounts of capital to scale up. So blocking distillation is already a valuable endeavor, but the selective release approach to doing so also gives labs a way to differentiate their business offerings as the category becomes key to profitable implementation.

Whether Mythos or any new model actually poses a threat to the security of the internet remains to be seen, and careful rollout of the technology is a responsible way forward.

Anthropic did not respond to our questions at the time of writing about whether the decision also addresses distillation concerns, but the company may have found a smart approach to protecting the internet – and its bottom line.