Particle.news
Download on the App Store

Anthropic Probes Unauthorized Access to Mythos During Restricted Trial

The breach probe highlights the risk calculus around a model kept to select U.S. partners because Anthropic says it can find long‑hidden software flaws.

Overview

  • Anthropic said it is investigating unauthorized access to its Mythos AI through a third‑party contractor’s development system used for external vendors.
  • The company said it has found no evidence the incident went beyond that isolated vendor environment, which is separate from systems used for paying customers.
  • AFP reported this is the third internal security issue flagged at Anthropic in the past month.
  • Mythos remains in a closed test with a small group of major U.S. tech and finance partners, including Amazon, Microsoft, Apple, NVIDIA and J.P. Morgan, because the company is delaying a wider release.
  • Anthropic says Mythos can outpace human analysts by spotting long‑standing bugs in software, a claim that has drawn questions from banks and officials in the UK and EU, and criticism from some who see the warnings as marketing in a rivalry with OpenAI.