A White House official, speaking on the condition of anonymity regarding the upcoming meeting on Friday, indicated that the administration is collaborating with advanced AI labs to evaluate their models and the security of their software. The official emphasized that any new technology that might be utilized by the federal government will need a technical evaluation phase.
This meeting follows rising tensions between the Trump administration and Anthropic, a company focused on AI safety which aims to implement safeguards in AI development to minimize risks while maximizing its benefits for the nation’s security and economy.
President Donald Trump attempted to prohibit federal agencies from using Anthropic’s chatbot Claude, citing a contract dispute with the Pentagon, stating in a February social media post that the administration “will not do business with them again!”
Defense Secretary Pete Hegseth also moved to classify Anthropic as a supply chain risk, a unique action against a U.S. company, which Anthropic has challenged in two federal courts. The company seeks assurance that the Pentagon will refrain from utilizing its technology for fully autonomous weapons and surveillance of U.S. citizens. Hegseth stated that the company must permit any applications deemed lawful by the Pentagon.
In March, U.S. District Judge Rita Lin ruled against enforcing Trump’s social media directive that banned federal agencies from using Anthropic products.
Anthropic has opted not to comment on the meeting ahead of time.
The San Francisco-based company announced on April 7 that its new Mythos model is so “strikingly capable” that its use is being restricted to select customers due to its proficiency in identifying and exploiting computer vulnerabilities beyond that of human cybersecurity experts.
While some industry professionals have questioned if Anthropic’s assertions about excessive AI capabilities are merely marketing tactics, even some of its most vocal detractors agree that Mythos could signify a significant advancement in AI technology.
David Sacks, a notable critic of Anthropic and former White House AI and crypto czar, advised that the situation should be taken seriously.
“Anytime Anthropic is alarming people, we need to question, ‘Is this a strategy? Is this part of their Chicken Little narrative? Or is it genuine?’” Sacks remarked on the “All-In” podcast he co-hosts with other tech investors. “Concerning cybersecurity, I would credit them and assert that this appears to be a real concern.”
Sacks added, “It’s logical that as coding models evolve and become more advanced, their ability to detect bugs escalates as well. This means they can locate more vulnerabilities and potentially combine several vulnerabilities to create an exploit.”
The potential advantages and risks associated with the model have also garnered interest outside the U.S.
The UK’s AI Security Institute reported evaluating the new model and deemed it a “step up” from prior iterations, which were already advancing swiftly.
“Mythos Preview can exploit systems with inadequate security measures, and it is probable that even more models with such abilities will emerge,” the institute indicated in their report.
Anthropic is also in discussions with the European Union concerning its AI models, including advanced systems not yet available in Europe, according to European Commission spokesperson Thomas Regnier on Friday.
Axios was the first to report the upcoming meeting between Wiles and Amodei.
When Mythos was introduced, Anthropic announced its initiative Project Glasswing, collaborating with major tech firms such as Amazon, Apple, Google, and Microsoft, as well as other organizations like JPMorgan Chase, to safeguard critical software from the significant risks that the new model could pose to public safety, national security, and the economy.
“We are providing it to a select group of important global companies and organizations to help them identify vulnerabilities,” stated Jack Clark, Anthropic co-founder and policy chief, at this week’s Semafor World Economy conference.
Clark noted that while Mythos is ahead of its peers, it is not a “special model.”
“Other systems similar to this will emerge within a few months from various companies, and within a year to a year-and-a-half, we will see open-weight models from China possessing these capabilities,” he predicted. “Hence, the world must prepare for more advanced systems that will be prevalent.”
