Key Highlights
- Andy Jassy, Amazon’s chief executive, alerted Trump administration officials following the discovery that Anthropic’s Fable 5 could potentially facilitate cyber intrusions
- White House officials demanded Anthropic address the security flaws or impose restrictions; President Trump authorized complete international access prohibition
- Anthropic terminated access to both Fable 5 and Mythos systems for all users worldwide to meet export control requirements
- The company characterized the security gaps as “relatively basic” and noted comparable features are available in other publicly accessible AI platforms
- The restriction may negatively impact Anthropic’s planned stock market debut and provide advantages to competitors such as OpenAI
Andy Jassy, leading Amazon as CEO, engaged in direct communications with federal authorities, including Treasury Secretary Scott Bessent, regarding security vulnerabilities discovered in Anthropic’s Fable 5 system. Amazon’s research team employed targeted queries to extract information from the model that could potentially enable cyberattack execution.
The model’s protective mechanisms were designed to prevent such information disclosure. These discoveries were transmitted to White House personnel and security agencies, triggering deliberations on appropriate countermeasures.
Government representatives instructed Anthropic to remediate the vulnerabilities or discontinue the model’s availability. During Friday discussions between Anthropic’s chief executive Dario Amodei and administration representatives, some officials perceived the company as resistant to collaborating with federal security specialists on solutions.
Administration leadership determined that implementing foreign access restrictions represented the most effective risk mitigation strategy. President Trump authorized the measure, though he privately expressed apprehension about potential innovation delays.
Complete Deactivation of Premium AI Systems
In response to the directive, Anthropic deactivated both Fable 5 and its Mythos platform for every user globally — extending beyond just international clients — to guarantee adherence to newly imposed export regulations.
The organization maintained that the identified weaknesses were fundamentally basic in nature. Anthropic emphasized that alternative publicly distributed AI systems possess capabilities to retrieve identical information.
Cybersecurity expert Andrew Morris, who established GreyNoise Intelligence, examined Amazon’s research results. His assessment indicated that while Fable 5 demonstrated capability to detect software vulnerabilities across a minimum of four applications, no proof existed that researchers obtained the more hazardous functionality of transforming those vulnerabilities into operational attack tools.
Anthropic has consistently prioritized safety protocols as fundamental to its operations. The organization previously postponed expanded Mythos availability following White House guidance and maintains collaboration with a federal AI evaluation division prior to launching new systems.
Implications for Anthropic’s Future
The circumstances present significant challenges for Anthropic. The organization is advancing preparations for a possible public offering potentially scheduled for autumn this year.
With its premier systems unavailable, clients may migrate to alternative providers. OpenAI maintains its own cybersecurity-oriented model and has established ongoing dialogue with the Trump administration.
Frictions between Anthropic and White House officials have historical precedent. The Defense Department previously classified Anthropic as presenting security concerns, a determination the company contests through two concurrent legal proceedings.
National Cyber Director Sean Cairncross and Commerce Secretary Howard Lutnick participated in deliberations preceding the prohibition. The Commerce Department maintains authority over export regulations affecting critical technologies.
Commerce’s prohibition currently bars foreign governmental entities, corporations, and private individuals from utilizing Fable and Mythos. Numerous members of Anthropic’s research workforce hold foreign citizenship, which the company indicated essentially prevents their continued work on these systems.
White House AI policy adviser David Sacks characterized the restriction as implemented “reluctantly” and conveyed optimism that Anthropic would address the problems, enabling the models’ restoration to widespread availability.
