Topline
The U.S. lifted export restrictions on Anthropic’s Mythos models on Tuesday night, allowing the company to begin restoring users’ access to its most advanced new models, after the Commerce Secretary Howard Lutnick said the Trump administration worked with the AI firm to “analyze and approve” the model.
The U.S. government slapped Anthropic with export restrictions days after the Fable 5 model was released to the greater public over cybersecurity safety concerns.
NurPhoto via Getty Images
Key Facts
Anthropic launched Fable 5, a model based on its powerful Mythos model, earlier in June, noting that it included safety features that blocked users from inquiring about cybersecurity, as well as biology and chemistry needed to create bioweapons.
However, the Commerce Department quickly forced the company to pull the models after issuing an export restriction days later.
In a blog post on Tuesday, Anthropic provided more details about the security concerns that lead to the export restrictions in the first place, including confirming they came after the government “became aware” of a report from researchers at Amazon, who were able to successfully bypass Fable 5’s safety mechanisms and identify software vulnerabilities.In one instance, the Amazon researchers were also able to get Fable 5 to “demonstrate” how a software security vulnerability could be exploited.
However, the specific technique used by the Amazon researchers is now “blocked in over 99% of cases,” Anthropic said on Tuesday.
When Will the Mythos Models Be Available Again?
Mythos 5, the more powerful version of Anthropic’s latest model with fewer safeguards, was made available to what the company called a “small number of trusted Project Glasswing partners for use in defensive cybersecurity.” Project Glasswing consisted of a larger group of companies who received early access to the technology to help identify risks. Fable 5, the version with more safeguards, will be available to the general public starting again on Wednesday, Anthropic said.
Contra
Since Anthropic began offering glimpses of their Mythos model earlier this year, critics have slammed the company for allegedly building hype by talking up how dangerous the model could be in the wrong hands. Sam Altman, the CEO of OpenAI, one of Anthropic’s biggest rivals, implied the company was engaging in “fear-based marketing” in an appearance on the Core Memory podcast in April. Responding to a question about the headlines generated by Mythos earlier this year, Altman said, “it is, you know, clearly incredible marketing to say, “We have built a bomb. We’re about to drop it on your head. We will sell you a bomb shelter for $100 million.”
Key Background
Tuesday’s news could signal a fresh start for Anthropic, who have been at odds with the Trump administration for months over use of their AI models. In March, the Pentagon designated the company a supply chain risk following a spat between Defense Department officials and the AI firm, who refused to allow the government unrestricted access to its models. Anthropic has repeatedly stated it would not allow its models to be used for autonomous weapons or mass surveillance of Americans. The news also offers a positive signal as the company prepares for what could be a blockbuster IPO.