Amazon's Jassy Raised Security Flags on Anthropic Models Before Trump AI Export Curbs

Amazon CEO Andy Jassy was among several tech leaders who expressed specific concerns to senior Trump administration officials regarding security risks in Anthropic's Claude Mythos 5 and Fable 5 models, according to two sources familiar with the discussions. The conversations occurred in the days preceding the administration's announcement of new export controls targeting those particular models, suggesting Jassy's input may have influenced the policy timing.

As Amazon's largest external AI investment—a $4 billion commitment made in late 2023—and a key vendor relationship through AWS, Amazon holds significant leverage and insight into Anthropic's operations. The company has integrated Anthropic's models into its Bedrock service and uses them internally for various applications, giving Jassy's security assessments particular weight in policy circles. Sources indicate the concerns centered on potential misuse scenarios rather than technical flaws, though specific details were not disclosed.

The Trump administration's subsequent action added restrictions on exporting the Mythos 5 and Fable 5 models to certain foreign entities, citing national security risks under updated Export Administration Regulations. This marks one of the first instances where direct feedback from a major corporate investor precipitated specific AI export controls under the current framework, reflecting heightened scrutiny of foundation model proliferation.

Industry analysts note the development underscores the complex position of cloud providers like Amazon, which simultaneously fund, distribute, and rely on external AI innovation while navigating evolving government risk assessments. While Amazon benefits from its Anthropic partnership through AWS revenue and product enhancement, it also faces potential complications if export limits constrain model availability for its global customer base.

The episode adds to a pattern of increasing government intervention in AI development, following earlier semiconductor export controls and recent executive orders addressing AI safety. It also highlights how major tech companies' dual roles as investors and consumers of AI technology create unique channels for influencing policy—though critics argue such access risks privileging incumbent players over newer entrants in the AI ecosystem.

For further context on the Amazon-Anthropic relationship, see Amazon's September 2023 investment announcement. Official details on the export controls can be found in the BIS notice regarding AI-related transactions. Anthropic's public model documentation is available at their official website.