Dozens of U.S. lawmakers are urging the White House to move faster against AI‑driven cyber threats, warning that tools like Anthropic’s Mythos could supercharge hacking if left unchecked. In a new bipartisan letter, 32 members of Congress press the administration to treat advanced AI models as a serious national‑security issue, not just a tech story.
Mythos is Anthropic’s latest high‑end AI model, designed to hunt for software flaws at a speed and scale humans cannot match. Early tests reportedly showed it can uncover thousands of previously unknown vulnerabilities, raising alarms in banks, tech firms, and government agencies. Anthropic has limited access through a program called Project Glasswing and is only working with a small group of companies while defenses catch up.
Older models from Anthropic and OpenAI can already reproduce many of Mythos’ headline capabilities when carefully orchestrated. Mythos stands out because it not only finds bugs but also helps generate working exploits with far less human input. That possibility has made lawmakers and regulators worry about a wave of automated attacks on schools, hospitals, banks, and government systems.
What Congress Wants From the White House
According to Axios, the 32 lawmakers in their letter ask the White House to explain how it plans to manage “frontier” AI models that can penetrate critical software. They want clear rules on when companies must notify the government if a new model creates serious security or safety risks. They also push for stronger coordination between federal agencies and private firms that are already testing artificial intelligence on real‑world systems.
The letter comes as officials discreetly debate whether voluntary safety assurances are sufficient. Some politicians are also discussing licensing or greater monitoring of the most powerful models. Others worry that heavy-handed regulation could hinder valuable security research. That division sets up an early showdown over how far Washington should go to regulate AI capabilities used in cybersecurity.
The White House has previously assembled IT and cybersecurity professionals to talk about Mythos and similar tools, asking which codebases they’re analyzing and how they’re prioritizing updates. Anthropic and OpenAI, meanwhile, say they’re issuing early warnings about AI-enabled cyber threats and rolling out systems in phases to give defenders time to adjust. But security researchers say attackers, especially those supported by governments, already know what to do with the holes if they are discovered.
READ MORE: Bullish Stock Stalls Ahead of Earnings: What Next For BLSH?
