Particle.news
Download on the App Store

U.S. Weighs AI Model Reviews as Anthropic Restricts Mythos and Lands $1.5 Billion JV

The move signals a shift toward tighter oversight of powerful models that can find software flaws and build exploits.

Overview

  • The White House is considering a working group and pre‑release safety reviews for new AI systems, according to reports Tuesday, though a spokesman called talk of decrees speculative.
  • Anthropic’s Claude Mythos remains under restricted access through its Glasswing program after internal tests showed it could identify thousands of high‑risk bugs and autonomously craft a working exploit.
  • The company is investigating a report that some users reached Mythos through a third‑party environment, and it also suffered a separate source‑map error that exposed Claude Code’s source.
  • Anthropic announced a roughly $1.5 billion joint venture with Blackstone, Hellman & Friedman, Goldman Sachs and others to help companies deploy its tools and rework workflows.
  • Security agencies and experts, including Germany’s BSI, warn models like Mythos could help criminals target critical software and everyday services such as online banking.