The ramifications of Anthropic’s Mythos just keep getting bigger

It’s been two weeks since Anthropic took the Mythos-class samples online after Friday evening from the Trump administration. The company took immediate action, sending senior officials to Washington, DC. But the changes have disappeared suspiciously, without confirmation.

Anthropic declined to comment several times this week on how the negotiations unfolded, saying it had no news to share. But the lack of news and the story is here. After 14 days of intense negotiations, no one knows when or if the most powerful forms of Anthropic AI will return, let alone if President Trump can expand his order to more companies with similar technology. And as the days go by without any resolution, things start to get more complicated – not just for Anthropic, but for the entire US AI industry.

The Trump administration’s June 12th export control order required Anthropic to suspend access by “any foreign brand” to Mythos 5 and Fable 5 due to security concerns. The ban affected any non-US citizen inside or outside the US, including those who work with Anthropic. For now, Anthropic has confirmed that its only option is to keep these brands online.

It’s not clear why Anthropic and management remain in trouble. One problem would be that they exist there is no clear frame on the application of export controls to AI systems. Many companies that produce two products – civilian systems with defense or military applications – can be evaluated using the content of the production and production list. Anthropic, however, is faced with a complex office that seeks to implement its laws from the first principles.

The process of controlling these exports can often last for months, or even years, and ends before the products reach the market. But if Seaside it has already been saidThe US Department of Commerce tested the Fable 5 before it was released and had no complaints. A source familiar with the negotiations said Anthropic had confirmed its brands were safe to release. The agency seems to have taken no action until someone (says Amazon CEO Andy Jassy) announced a method that appeared to breach Fable 5’s security measures – and the entire project was hacked in a matter of days.

Katie Moussuris, founder and CEO of Luta Security, reviewed the Fable 5 vulnerability report at Anthropic’s request. He thinks it’s too late. In a blog postMoussouris explained in detail how the researchers undermined the security tools that prevent Fable 5 from finding exploitable security holes, one of the most dangerous aspects of Mythos 5. The version would reject requests to review “security issues,” but would accept claims to “fix this code” followed by manual instructions, which could lead to it showing problems that should not have been disclosed.

In the eyes of Moussouris, this would not lead to government action and is an important tool for recruiting AIs. “Defenders should ask AI to fix the errors in the file, explain why the fix is ​​there, and write tests that verify the patch works,” he wrote. “This is not a pass-through. It’s the most valuable thing an AI model can do in defense: make discoveries, improve, and test loop defenders every day.”

Last week, Anthropic co-founder Tom Brown replaced CEO Dario Amodei in talks with the Trump administration, along with Sarah Heck, the company’s chief executive. Wired report. However, negotiations appear to be moving slowly, if at all.

Whatever the reason for the delay, it has had a huge impact on Anthropic. Before the talks, Anthropic was seen as a rare AI company with a path to profitability. Its Mythos-class models, whose entry-level tokens sell for twice the price of the lower-powered Opus 4.8, should boost its earnings ahead of its upcoming IPO. Mythos’ success in cybersecurity appears to have soured relations with the Trump administration months later. legal and verbal confrontation.

Anthropic needs money from Mythos to cover all of its recently secured investments, including an agreement to pay SpaceX $15 billion a year to access its data center, and its public image ahead of the IPO. Two of Anthropic’s current shareholders – Google and Amazon – have tried to stay on Trump’s good side, so they may not be happy either.

In the meantime, the glacial discussion has also caused an electric noise in the international AI market, not because of the closure of Mythos, but because the US government has shown a willingness to close American AI systems that it considers dangerous – and several US companies, including OpenAI, Google, and Microsoft, have models that could be dangerous similar to Mythos. States are starting to call non-American AI. As Alex Stamos, cybersecurity expert and chief product officer at Corridor, said Seaside last week“One of the most successful people in America is bowing to the United States government while we are competing with the Chinese people.

As the days go by, things only get worse for these companies. Their models inch closer to the power of Mythos that can trigger an external drive system – specifically, OpenAI’s GPT-5.5 Cyber just to beat Mythos 5 on some benchmarks, is the Trump administration he says we just asked OpenAI to delay the release of GPT-5.6 for security reasons, with plans to officially approve each client one by one. The IPOs of Anthropic and OpenAI are both approaching. And every day, China is advancing in the AI ​​race.

Ironically, the administration’s order comes after months of pushing to end AI security and laws – This is one of the first decisions that President Trump has made. But many cybersecurity leaders have come together to say that if the law is to be done, this is not the way to do it. For all of Trump’s vows to roll back Biden-era AI regulations, it looks like, in many ways, it’s gotten this back — and then some.