Overview
- Cybersecurity stocks slid Friday after reports that Anthropic is testing a new model called Claude Mythos that could turbocharge hacking, with ETFs and names like CrowdStrike and Palo Alto Networks falling several percent.
- Fortune found a publicly accessible cache tied to Anthropic’s website system that exposed nearly 3,000 unpublished files, and the company said human error in a content management tool made draft materials reachable without a login.
- Anthropic said the drafts did not include core infrastructure or customer data and confirmed it quickly locked down access once notified, stressing the issue was unrelated to its Claude AI tools.
- The leaked drafts described Mythos as a step change that outperforms Claude Opus 4.6 in coding, reasoning, and cybersecurity, introduces a larger Capybara tier above Opus, costs more to run, and is in early testing with select customers focused on cyber defense.
- Coverage varied by outlet, with Fortune first reviewing the cache and its risks, Quartz corroborating the scale and Capybara naming, and CNBC zeroing in on market moves, while separate reports said Anthropic is exploring an IPO as soon as October to fund heavy compute needs.