Anthropic still pitches Mythos to Trump administration

Anthropic is discussing its new AI model Mythos with the Trump administration despite the Pentagon cutting off business and barring its use by the Pentagon and its contractors, tied to a dispute over guardrails for military use, co-founder Jack Clark said in Washington.

Clark said the dispute was narrow and that Anthropic would keep talking with the government on Mythos and future models.

The agencies involved were not clear, and Anthropic has said Mythos is its most capable yet for coding and agentic tasks, which can let software act with more autonomy.

Experts said strong coding ability can help identify cybersecurity flaws and suggest ways to exploit them, and a federal appeals court recently let the Pentagon blacklist stay in place for now.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

The dispute centers on two specific AI red lines

Anthropic wants Pentagon contracts to spell out bans on using its AI for “mass domestic surveillance of Americans” and for “fully autonomous weapons” 1.
The Pentagon declined and asked for permission to use the technology for “all lawful purposes,” saying existing laws and Defense Department policies already block the uses Anthropic worries about 1.
The deadlock stands out because OpenAI said it struck a Department of Defense deal with safety principles that prohibit domestic mass surveillance and require human responsibility for the use of force, including for autonomous weapon systems 2.
Claude, Anthropic’s AI model, has already been used by the U.S. military in combat operations, including strikes on Iran, though its exact role remains unclear 1.

Mythos heightens cybersecurity concerns and may compress patch timelines

Anthropic describes Mythos’s knack for finding security flaws as an emergent result of better code skills, reasoning, and autonomy rather than a capability it trained directly 3.
Anthropic says Mythos Preview can spot then exploit zero-day vulnerabilities, meaning previously unknown security flaws, across every major operating system and every major web browser it was directed at 3.
Commentators argue this could shrink the time between disclosure and weaponization, which raises pressure on companies and government agencies whose patch cycles can take weeks 4.
Anthropic is restricting Mythos Preview through Project Glasswing, a controlled program for select partners, which has sparked debate over who should receive access to such powerful tools 5, 6.