March 27 was probably a pretty strange day for Anthropic. In the morning, the world learned that the company behind the Claude AI assistant was seriously considering going public. In the afternoon, news broke that somewhat complicated the celebration: the company had accidentally published thousands of internal documents, including details about its as-yet-unannounced and most powerful AI model ever. They call it Claude Mythos.
Information leak
The mistake was trivial. The content management system operator forgot to mark the materials as private. And suddenly, there were more than 3,000 unpublished files on the public internet: draft blog posts, internal documents, and details about a planned private event for European business leaders at which CEO Dario Amodei was due to speak in person.
But the biggest bombshell was another file. It was a draft blog post describing a model that Anthropic also referred to in its internal materials as "Capybara", a new category above the existing Opus series models. Both names, Mythos and Capybara, apparently refer to the same model.
Anthropic later acknowledged the mistake. A company spokesperson confirmed that the model does indeed exist and is being tested. "We are developing a general-purpose model with significant advances in reasoning, coding, and cybersecurity," a company representative said.
What do we know about the Mythos model?
The leaked documents describe Mythos as "by far the most powerful AI model we have ever developed." Compared with Claude Opus 4.6, it reportedly achieves dramatically higher scores in programming, academic reasoning, and cybersecurity tests. And cybersecurity is precisely the part that literally unsettled the markets. The leaked draft claims that the model is "far ahead of any other AI model in cyber capabilities" and that it may be capable of exploiting software vulnerabilities faster than companies can patch them.
It is therefore no surprise that shares of companies such as Palo Alto Networks, CrowdStrike, and Fortinet fell by 5 to 9 percent on Friday. The market worked out what it could mean if such a tool fell into the wrong hands. Anthropic responded by announcing that Mythos will not be available to the general public. At least not yet. Only selected customers focused on cyber defense will be given access, meaning companies that will use it to protect systems, not attack them.
Planned IPO
Yet the entire day was supposed to be about a different story. Both Bloomberg and The Information reported that Anthropic was considering going public as early as the fourth quarter of 2026, specifically in October. The company had reportedly held preliminary talks with Goldman Sachs, JPMorgan, and Morgan Stanley.
It is a huge bet. Anthropic was valued at approximately $380 billion in February 2026, and an IPO could raise more than $60 billion. For comparison, the company was only founded in 2021, when co-founders Dario and Daniela Amodei launched it with $2.5 billion. Such a leap in five years is far from common.
But going public just as the controversy over the leaked documents is gathering momentum? And on top of that, there is the legal dispute with the U.S. Department of Defense, which labeled Anthropic a "supply chain risk" after the company refused to allow the unconditional military use of its models. Federal Judge Rita Lin did grant the company preliminary protection and described the government's actions as "Orwellian," but the case is still ongoing.
Technology stocks had a rough week even without Anthropic. Meta and YouTube were fined by a Los Angeles court for harm caused to young users by the platforms' addictive design. Meta shares fell by more than 7 percent. Reddit and Snap dropped by more than 10 percent.
On top of all that came pressure from falling memory chip stocks after Google unveiled its new compression algorithm, TurboQuant, which reduces the memory requirements of AI models. Micron, SanDisk, and SK Hynix all took a beating. Then came Anthropic with its leaked model and IPO plans.
Additional sources: ca.finance.yahoo.com and thenews.com
