Amazon CEO Andy Jassy raised security and safety concerns about Anthropic's AI models before the company restricted global access to Claude 3.5 Sonnet and Claude 3.5 Haiku on Friday, according to reporting from TechCrunch. The timing suggests Jassy's intervention may have influenced Anthropic's decision to pull the models from worldwide availability.
Anthropic did not publicly explain the removal, leaving the AI safety community searching for answers. The restriction affects both the web interface and API access for these two versions of Claude, though the company left earlier model versions available to users.
Jassy's concerns centered on security and safety implications of the models, sources told TechCrunch. The nature of those specific concerns remains unclear, but they apparently carried enough weight to prompt swift action from Anthropic leadership. Amazon holds a significant stake in Anthropic, having invested up to $4 billion in the startup, giving Jassy considerable influence over the company's direction.
The move comes amid broader government scrutiny of AI safety practices. Several regulatory bodies have increased oversight of large language model deployments, particularly around potential misuse and security vulnerabilities. Anthropic has positioned itself as safety-focused, so flagged concerns from major stakeholders like Jassy may align with the company's existing risk management philosophy.
What remains unclear is whether Jassy acted independently or as a representative of government interests. The lack of transparency from both Amazon and Anthropic fuels speculation about what triggered the restriction. Neither company has issued detailed public statements explaining the decision.
The incident highlights the complex relationship between major AI investors and developers. When a stakeholder like Amazon raises concerns, startups face pressure to respond quickly, potentially without full public disclosure. For Anthropic users relying on these models for production systems, the sudden availability loss creates operational uncertainty and raises questions about communication protocols during potential security