Anthropic leak rattles tech and crypto — new “Claude Mythos” model exposed
A leaked trove of internal materials from Anthropic, the AI firm behind the Claude models, briefly rattled both cybersecurity stocks and the crypto market after details surfaced about a more powerful model in testing, Fortune reported.
What happened
- Roughly 3,000 assets tied to Anthropic’s blog were accidentally exposed in a publicly accessible data store, revealing draft announcements and internal documents that had not been published.
- One draft blog post named the new system “Claude Mythos” and framed it as “a step change” in capability — “the most capable we’ve built to date,” Anthropic says — currently being trialed with a small group of early access customers as the company evaluates behavior and risks.
Why markets moved
- The leaked materials included warnings that the model could pose serious cybersecurity risks, including an ability to identify and potentially exploit software vulnerabilities. That alarm likely pressured investors in security-focused firms: Palo Alto Networks (PANW), CrowdStrike (CRWD) and Fortinet (FTNT) all slid 4%–6%, while the iShares Expanded Tech-Software ETF (IGV) fell about 2.5%.
- Crypto markets also reacted. Bitcoin, which had flirted with $70,000 earlier, dropped back toward the mid-$60,000s — trading around $65,983.92 when the pullback occurred — with traders pointing to the overnight leak as a contributing factor.
What’s next for Anthropic
- Anthropic currently markets three model tiers — Haiku, Sonnet and Opus — varying by size, cost and capability. The leaked files suggest a new, even larger tier called “Capybara,” positioned above Opus in scale and intelligence.
- Anthropic is publicly testing the newest model with a limited set of partners while assessing its behavior and associated risks.
Why crypto watchers should care
- Advanced models that can surface and explain software vulnerabilities raise both productivity and security questions for crypto infrastructure — exchanges, custodians and smart contracts increasingly depend on fast AI-driven tooling, but those same tools could accelerate exploit development if misused. The incident underscores how developments in cutting-edge AI can ripple quickly through cybersecurity markets and digital assets.
Fortune broke the story; Anthropic provided the characterization of the model and said it is evaluating risks as it expands testing.
Read more AI-generated news on: undefined/news
