The Digester
general

Anthropic left powerful new AI model exposed online, raising cybersecurity alarms

Nearly 3,000 internal Anthropic assets were publicly accessible, revealing plans for a highly capable AI called Claude Mythos and prompting security warnings as markets, bond yields and oil prices react to rising geopolitical risk.

Mar 27th 2026 · United States

Insights

  • Anthropic unintentionally exposed about 3,000 internal assets on its website, including files on an in-development model called Claude Mythos.
  • Company documents warn Claude Mythos will be Anthropic's most capable model and that it could pose unprecedented cybersecurity risks.
  • A US judge blocked the Pentagon from labeling Anthropic a supply chain risk to bar Claude from government work.
  • US markets fell, with the S&P 500 down 1.74% and the Nasdaq down 2.38%, while S&P futures were down 0.14% ahead of the open.
  • European sovereign yields spiked, with 10-year German bunds closing at 3.07% and 10-year French OATs at 3.80%.
  • Brent crude topped $110 per barrel as disruptions around the Strait of Hormuz pushed tanker costs higher and added roughly $14 per barrel to shipping costs.
  • President Trump said he would give Iran 10 more days before ordering destruction of its energy infrastructure, and the Pentagon is considering sending about 10,000 additional troops to the Middle East.

Sources