Amazon CEO Andy Jassy reportedly flagged security vulnerabilities in Anthropic's AI models to government officials, prompting the AI safety startup to restrict global access to two of its models on Friday.
The move marks a rare moment of tension between two companies that have maintained a close partnership. Amazon has invested over $4 billion into Anthropic and hosts much of the startup's infrastructure through AWS. Jassy's concerns centered on potential security risks within the models themselves, according to sources familiar with the matter.
Anthropic, founded by Dario and Daniela Amodei, responded swiftly by taking Claude 1.3 and Claude Instant offline for users outside the United States. The restriction affects a significant portion of Anthropic's global user base and represents a dramatic escalation from typical product updates or bug fixes.
The timing is notable given ongoing regulatory scrutiny of large language models and AI safety practices. Jassy's intervention with government officials suggests the concerns went beyond routine technical feedback that would normally flow through Amazon's partnership channels with Anthropic. This path indicates either severe security implications or a broader policy-level disagreement about model deployment practices.
Amazon's substantial stake in Anthropic makes this move particularly striking. The e-commerce and cloud giant views the AI startup as a strategic asset in its competition with OpenAI and Google. Yet Jassy apparently deemed the security issues serious enough to escalate beyond standard partnership protocols.
Anthropic has not publicly detailed the specific vulnerabilities that triggered the worldwide access cutoff. The startup built its reputation on AI safety research and constitutional AI principles, which may explain the swift action despite potential business impact.
The incident underscores growing tension within the AI industry over model safety, deployment practices, and government oversight. It also reveals that even major strategic partnerships can fracture when security concerns emerge. For Anthropic, the move protects its safety