Just weeks after a sweeping government ban, Anthropic, a leading artificial intelligence firm, found itself back in the heart of the White House. A high-stakes meeting between White House Chief of Staff Susie Wiles and Anthropic CEO Dario Amodei signaled a dramatic shift in perspective, raising questions about the administration’s approach to powerful AI technology.
The reversal centers on Anthropic’s newest creation, Mythos Preview – an AI model so advanced, the company itself has tightly controlled its release. Concerns about potential misuse prompted this limited access, but its capabilities are now being re-evaluated in light of national security needs.
This sudden reconsideration comes after a sharp break with the Pentagon. Defense Secretary Pete Hegseth designated Anthropic a national security “supply chain risk,” effectively barring its technology from military applications. The company is now fighting that designation in court, arguing it’s unlawful and retaliatory.
The initial dispute stemmed from Anthropic’s refusal to grant unrestricted access to its AI. The company insisted its systems not be used for mass surveillance or autonomous weapons, a condition Pentagon officials deemed unacceptable. This led President Trump to order a government-wide halt to the use of Anthropic’s models.
Now, officials are grappling with a critical question: could the very system previously flagged as a risk be essential for defending U.S. infrastructure? The internal debate highlights the complex challenge of harnessing AI’s potential while mitigating its dangers.
Adding another layer to the situation are past political clashes. Amodei himself has been critical of Trump, once comparing him to a “feudal warlord” in a social media post. Internal company communications, leaked earlier this year, suggested the administration’s actions were partly fueled by a lack of “dictator-style praise.” Amodei later apologized for the tone of the message.
Anthropic’s governance structure is also under scrutiny. The company is overseen by a “Long-Term Benefit Trust,” giving non-financial stakeholders – including national security and policy experts – significant influence over its direction. Figures like Neil Buddy Shah and Richard Fontaine are among the trustees.
The company’s financial backing further complicates the picture. Early investors included prominent Democrats like Dustin Moskovitz and Eric Schmidt, as well as Sam Bankman-Fried’s FTX. However, major tech giants like Amazon, Google, and Microsoft have also invested, reflecting Anthropic’s growing importance in the global AI landscape.
Anthropic has actively sought to bridge the political divide, hiring staff with Republican backgrounds and adding Chris Liddell, a former Trump White House official, to its board. The company has also contributed to a bipartisan group supporting AI regulation.
Despite these efforts, internal criticism persists. White House AI advisor David Sacks has accused Anthropic of attempting “regulatory capture,” using safety concerns to gain a competitive advantage. Anthropic vehemently denies these claims.
Mythos Preview’s potential is undeniable. Currently available through a limited program called Project Glasswing, the system has already identified thousands of previously unknown security flaws in critical systems – some dating back decades. This highlights both its defensive capabilities and the potential for misuse.
Anthropic warns that the rapid advancement of AI means such powerful tools will inevitably proliferate, potentially falling into the wrong hands. The consequences for economies, public safety, and national security could be severe, underscoring the urgency of navigating this complex technological landscape.
