Anthropic CEO Says AI Could Be as Dangerous as Nuclear Weapons — Says the Law Needs to Catch Up

TLDR

• Anthropic CEO Dario Amodei warned AI could become as dangerous as nuclear weapons
• Anthropic released a formal proposal calling for mandatory AI safety testing and third-party audits
• The proposal asks government to have power to block AI models deemed too risky
• Congress faces long odds — under 30% chance — of passing any AI legislation this year
• Anthropic is currently suing the Trump administration over Pentagon restrictions on its products

---

Anthropic, the company behind the Claude AI, has released a set of proposals calling on Congress to pass new AI safety laws. CEO Dario Amodei says the technology is advancing fast enough to pose serious risks to national and global security.

In a blog post, Amodei compared powerful AI models to nuclear materials, writing that they could soon look “less like airplanes or automobiles and more like weaponizable nuclear materials.”

He called on the government to treat AI regulation the way the Federal Aviation Administration handles airplane safety — with mandatory testing before any public release.

What Anthropic Is Proposing

Anthropic’s proposal asks Congress to require AI companies to submit their most powerful models to third-party audits. Those audits would check for risks in four areas: cybersecurity, biological weapons, loss of control of AI systems, and automated research and development.

The proposal also calls for companies to report “safety incidents” and gives the government the ability to block or pull back any model that fails to meet safety standards.

This puts Anthropic at odds with much of the AI industry, which has pushed back against what it sees as government overreach.

President Trump signed an executive order earlier this month that set up a voluntary review process for new AI models. Anthropic said it welcomes that step but is pushing for it to go further and become mandatory.

Legislation Faces Long Odds

Two lawmakers — Rep. Jay Obernolte and Rep. Lori Trahan — unveiled a draft bill last week that would require AI companies to create risk mitigation plans. Sen. Ted Cruz has also said he wants to move an AI bill forward.

But both efforts face opposition. Some advocates worry the bills would block stronger state-level rules from taking effect.

Stifel analyst Brian Gardner put the odds of any AI bill passing at less than 30%. Congress has roughly two months before its August recess, and lawmakers are already dealing with a crowded legislative calendar.

A digital assets bill, a housing bill, and a surveillance law reauthorization are all competing for floor time. Trump has also told Congress to prioritize a military funding bill and a voter ID measure.

Anthropic’s relationship with the federal government has been rocky. The Pentagon labeled the company a supply chain risk and banned the use of Claude across the department.

Anthropic sued the Trump administration, challenging that designation and a directive stopping civilian agencies from using its products. The dispute started after talks broke down over whether Claude could be used in autonomous weapons and mass surveillance programs.

Anthropic had insisted its technology not be used in fully autonomous lethal weapons. The Pentagon said it needed access to Claude for “all lawful uses.”

Weeks after filing the lawsuits, Anthropic released Mythos, its most advanced AI model, which drew fresh attention from the White House over potential safety risks.