Anthropic faces 2nd security scare as hackers claim Mythos breach

Hacker group ShinyHunters claims it has breached internal systems of AI-safety leader Anthropic, allegedly accessing data related to its unreleased Claude Mythos model. The unconfirmed claim follows a separate breach in July where a Discord group gained access to the same high-risk model.

Anthropic has not yet publicly confirmed the breach claim from ShinyHunters. The company previously stated it was "aware of the claim and investigating" the July incident, according to a Bloomberg report.

ShinyHunters allegedly shared screenshots of user management panels and internal AI experiments. This follows the July breach where users guessed the model's online location and used a contractor's privileged access. Anthropic has described Claude Mythos as capable of "identifying and then exploiting zero-day vulnerabilities in every major operating system."

The repeated security lapses, if the new claim is verified, could severely damage Anthropic's reputation as a safety-focused AI lab and erode trust among its enterprise customers and partners like Google. The incidents raise critical questions about security protocols across the entire AI industry as companies race to develop increasingly powerful models.

Second Security Incident in a Month

The claim from ShinyHunters marks the second time in recent weeks that unauthorized users have allegedly gained access to what Anthropic calls a "potentially dangerous" AI. In early July, a group of users on the social platform Discord gained access to Mythos not through a sophisticated attack, but by guessing its online location based on the company's naming conventions.

That earlier breach was aided by a member of the group who had privileged access as a contractor for Anthropic. While the users told Bloomberg they were using the model for harmless tasks, the incident exposed significant security weaknesses. For a company whose brand is built on safety and responsible AI development, two alleged breaches of its most powerful model in quick succession is a serious concern for investors and customers.

What is Claude Mythos?

Anthropic has kept Claude Mythos under tight wraps, granting access only to a select group of partners through an initiative called Project Glasswing. The company's own description of the model's capabilities—that it can find and use novel software vulnerabilities—places it in a unique risk category. While rivals like OpenAI and Google are also developing powerful models, Anthropic's public emphasis has consistently been on containing the potential dangers of advanced AI.

The alleged breaches suggest that the company's own operational security may not match its high-minded safety principles. If a group like ShinyHunters can access internal dashboards, it raises the possibility of model theft or the exposure of sensitive data from Anthropic's partners. The incidents provide a stark reminder that as AI models become more powerful, the companies building them become more valuable targets.

This article is for informational purposes only and does not constitute investment advice.