Anthropic Leak Exposes Claude Mythos With “Unprecedented Cybersecurity Risks” as Stocks Slide and Internal Documents Surface

Internal Materials Reveal Model Capabilities, Limited Rollout Plans, and Market Reaction Across Cybersecurity Sector

TL;DR

• Leak reveals Claude Mythos as Anthropic’s most powerful AI with explicit cybersecurity risk warnings
• Internal documents expose testing status, roadmap including Capybara, and invite-only executive summit plans
• Cybersecurity stocks drop sharply following leak, with major firms posting losses up to 7%

We’ve launched the all-new COIN360 Perp DEX, built for traders who move fast!

Trade 130+ assets with up to 100× leverage, enjoy instant order placement and low-slippage swaps, and earn USDC passive yield while climbing the leaderboard. Your trades deserve more than speed — they deserve mastery.

Anthropic is developing a new artificial intelligence system called Claude Mythos, described in internal materials as its most capable model to date, with advances spanning reasoning, coding, and cybersecurity. The company characterized the system as a “step change” in capability, with leaked documents indicating the model is already undergoing testing. The materials describe Mythos as a general-purpose system positioned above existing Claude models, including Opus, and designed to significantly expand performance across technical domains including software engineering and security operations.

The leak originated from a misconfigured content management system that exposed internal assets to public indexing, allowing access to nearly 3,000 unpublished files before the issue was identified and resolved. The exposed materials included draft blog posts, PDFs, images, and audio files, offering detailed insight into Anthropic’s product roadmap and internal risk assessments. The documents remained accessible until Anthropic was alerted, after which the company removed public access to the cached data.

Internal disclosures explicitly warned that Claude Mythos presents “unprecedented cybersecurity risks,” citing its ability to identify and exploit vulnerabilities at a level exceeding existing models. One internal statement read, “Although Mythos is currently far ahead of any other AI model in cyber capabilities, it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.” The documents described the model’s cyber capabilities as exceeding competing systems while raising concerns about defensive readiness.

Anthropic outlined a controlled deployment strategy, stating that Mythos will be released initially to a limited group of users, prioritizing organizations focused on cybersecurity defense. The company indicated that early access would be restricted to allow defenders to adapt systems and workflows before broader rollout. The materials emphasized that the phased release is designed to mitigate risks associated with advanced offensive cyber capabilities embedded in the model.

The leaked roadmap places Claude Mythos above the company’s Opus line, with internal references to future iterations under development, including a version identified as Capybara. The documents describe Mythos as larger and more intelligent than Opus, with testing results showing performance gains across software coding, academic reasoning, and cybersecurity tasks. Internal evaluations stated the model scored “dramatically higher” than Claude Opus 4.6 across these benchmarks.

The exposure also revealed details of an invite-only executive summit planned by Anthropic, where CEO Dario Amodei was expected to host business leaders at an 18th-century manor in the English countryside. The event was designed to showcase upcoming AI developments and discuss enterprise adoption strategies, according to the internal materials. The summit details were included among the unpublished assets made accessible through the leak.

Anthropic’s earlier disclosures tied its technology to real-world security incidents, stating that a Chinese state-sponsored group had used Claude Code in campaigns targeting about 30 organizations, including technology companies, financial institutions, and government agencies. The reference appeared in the leaked materials alongside broader warnings about the misuse potential of advanced AI systems in cyber operations.

Market reaction followed swiftly after the leak surfaced, with shares of major cybersecurity firms declining during trading. Palo Alto Networks dropped about 7%, CrowdStrike fell roughly 6.4%, Zscaler declined around 5.8%, and Fortinet decreased about 4%. The sell-off occurred as investors reacted to the implications of increasingly capable AI systems entering domains traditionally dominated by cybersecurity providers.

The developments follow a prior market event tied to Anthropic’s February release of Claude Cowork, which triggered a broad repricing across software and professional services sectors and erased roughly $285 billion in market value. Nexatech Ventures founder Scott Dylan commented at the time, saying, “The market's response was a signal, not that AI agents will immediately replace these businesses, but that investors are finally pricing in the structural risk that foundation model providers can now compete directly with the software layer,” adding that the same dynamic could extend into finance, procurement, and HR.

Public reaction included a statement from Elon Musk, who wrote, “Seriously troubling,” after details of the leak circulated. The comment appeared amid ongoing competition between AI firms and expanding commercialization efforts across the sector.

This article has been refined and enhanced by ChatGPT.