Overview
- Cybersecurity shares slid Friday after reports of an exposed Anthropic draft, with Palo Alto Networks, CrowdStrike, Okta and Zscaler down more than 5% and the Global X Cybersecurity ETF off as much as 6.1%, erasing about $14.5 billion in value.
- The exposed draft described an unreleased AI model called Claude Mythos, part of a Capybara line, and claimed unusually strong performance on reasoning, coding and cybersecurity tasks.
- Anthropic said a human error in its content system made nearly 3,000 assets accessible and a spokesperson told Fortune the exposure was unrelated to Claude, Cowork or any Anthropic AI tools.
- The leaked document, as reported by multiple outlets, indicated the company would delay a public launch and planned to share test results with security vendors to help them prepare, which has not been formally announced.
- Some market voices called the selloff an overreaction and pointed to ongoing growth at incumbents, including Palo Alto Networks’ rising revenue and early uptake of its AI security service Prisma AIRS, with company earnings next in focus.