Key Takeaways
- The White House gave Anthropic less than 90 minutes to suspend access to its newest AI models, Fable 5 and Mythos 5, citing national‑security concerns.
- Employees were left confused, receiving shifting explanations—first about foreign‑company access, then about a discovered vulnerability.
- Anthropic’s leadership met repeatedly with Trump‑administration officials, but no resolution emerged after six days of talks.
- The dispute follows an earlier clash over a $200 million Defense Department contract and a “supply chain risk” label that Anthropic is challenging in court.
- A cybersecurity paper from Amazon researchers suggested a flaw in Fable 5 that could be used to expose software vulnerabilities, prompting further scrutiny.
- Independent experts defended the model’s capability, arguing that prompting AI to identify and fix bugs is a defensive strength, not a weakness.
- More than 150 cybersecurity professionals signed an open letter urging the administration to lift the restrictions, claiming Anthropic was being unfairly targeted.
Background of the White House Directive
On a Friday morning, Anthropic executives received an urgent call from the White House: they had under 90 minutes to take down their newest AI models, Fable 5 and Mythos 5, because of national‑security worries. The directive arrived amid heightened scrutiny of advanced AI systems, and the company was told to disable access for all foreign nationals immediately. As one internal chat captured, an employee asked, “What are you telling your clients?” while another replied, “Does anyone know what to believe?” The abrupt timing left little room for preparation, forcing managers to scramble to notify customers of a possible service disruption.
Employee Reaction and Internal Confusion
Inside Anthropic, the news sparked a flurry of messages in employee group chats. Managers were told to brace for disruption, but the rationale kept shifting—first citing the risk of foreign companies gaining access, later pointing to a newly discovered vulnerability in the models. “I don’t understand what the issue is,” one worker confessed, reflecting the widespread bewilderment. The uncertainty extended to the company’s imminent IPO plans, with engineers wondering whether the White House order would jeopardize a public offering this year. The lack of clear, consistent information left staff in a holding pattern, repeatedly asking, “What should we believe?”
Leadership Engagement with the Administration
Over the following six days, CEO Dario Amodei and senior lieutenants held multiple meetings with Trump‑administration officials to resolve the standoff. Despite discussions on Monday and Tuesday, no breakthrough emerged; the U.S. order limiting access to Fable 5 and Mythos 5 remained in place. Anthropic issued a statement pledging its “ongoing commitment to working alongside the administration,” while continuing to seek clarification. The repeated engagements underscored the high stakes for both parties, yet also highlighted the administration’s reluctance to divulge the specific national‑security grounds for the directive.
Historical Tensions with the Trump Administration
This episode is not Anthropic’s first clash with the White House. Earlier in the year, the company disputed a $200 million Department of Defense contract for AI in classified systems, leading Defense Secretary Pete Hegseth to label Anthropic a “supply chain risk”—a designation never before applied to an American firm. Anthropic sued the government over that label, and the dispute flared again in April when it unveiled Mythos, a model touted for its ability to spot software vulnerabilities. The White House responded by discussing an executive order that would require voluntary pre‑release reviews of new AI models, a process in which Anthropic participated.
Release of Fable 5 and the Amazon Research Paper
Anthropic later released a “straitjacketed” version of Mythos, dubbed Fable 5, asserting that added guardrails made it safe for broad use. The model was provided to the Commerce Department for testing, with no explicit prohibition against deployment. Shortly thereafter, Amazon researchers published a preliminary paper describing a perceived security shortcoming in Fable 5. Using that alleged weakness, Amazon demonstrated how the model could be prompted to reveal flaws in specific software code. The findings were shared with Anthropic and then raised in a previously scheduled call between Amazon and administration officials, prompting the Treasury secretary to request a deeper briefing.
Government and Expert Assessments of the Findings
U.S. officials who reviewed the Amazon document characterized the results as “scary.” Yet cybersecurity experts countered that the ability to ask an AI to identify and explain software bugs is a valuable defensive tool, not a pathway to malicious hacking. Katie Moussouris, a noted cybersecurity authority, wrote in a blog post that “Defenders need to be able to ask A.I. to fix the bugs in a file, explain why the fix matters and write tests that confirm the patch works… That is not a guardrail bypass. It is the most valuable thing an A.I. model can do for defensive security.” She argued that restricting such capability would undermine defenders’ ability to secure their systems.
Broader National‑Security Concerns Cited by Officials
Beyond the Amazon paper, a government official told reporters that the White House’s apprehensions also involved unspecified national‑security worries about the companies Anthropic chose to partner with. This angle was never directly communicated to Anthropic, according to three individuals familiar with the discussions. The lack of transparency fueled speculation among employees that the administration’s actions might be motivated by factors other than technical risk, prompting one worker to ask, “Are we being bullied based on bad vibes?”
Support from the Cybersecurity Community
In response to the restrictions, more than 150 cybersecurity professionals signed an open letter to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross, urging the administration to lift the limits on Anthropic’s models. The letter emphasized that Fable 5 incorporates “multiple protections… to prevent its use for cyber offensive uses,” noting that those safeguards had become a source of humor in the cyber community upon launch. Signatories included luminaries from internet security, an AI expert at Nvidia, and a former NSA official responsible for responsible AI use. At Anthropic, employees circulated the letter as validation that they were being unfairly targeted, while also questioning the firm’s long‑term prospects if the White House continued to curb model releases.
Current Outlook and Unanswered Questions
As of early this week, Anthropic’s roughly 3,000 employees still lack a clear explanation for the White House order, and the company remains in negotiations with administration officials. The episode underscores the growing tension between rapid AI innovation and national‑security oversight, especially when advanced models possess dual‑use capabilities that can both defend and, potentially, offend. Until the administration provides concrete, transparent justification—or lifts the restrictions—Anthropic’s staff will continue to operate under a cloud of uncertainty, wrestling with both technical ambitions and the implications of geopolitical scrutiny.
https://www.nytimes.com/2026/06/17/technology/anthropic-trump-administration-fable.html