Overview
- Trump said the government should have a way to shut down dangerous AI and called the technology both a risk to bank stability and a way to make banking safer.
- His comments followed warnings about Anthropic’s unreleased Claude Mythos, which can hunt software bugs and could also help plan complex hacks against older banking systems.
- Anthropic has limited Mythos to about 40 vetted partners to find flaws in their code, and the company declined to expand on the specific warnings raised by experts.
- The U.S. government has labeled Anthropic a supply chain risk and barred federal use of its tools, and Anthropic is fighting that designation in court.
- Competitors are moving fast, with OpenAI developing GPT 5.4-Cyber to spot software weaknesses, while Anthropic says Mythos found a 27-year-old OpenBSD bug and thousands of likely zero-days that outside researchers have not yet verified.