Amazon CEO Andy Jassy reportedly alerted US government officials regarding security risks posed by Anthropic‘s Claude Fable 5 model before a global export ban was enacted. According to The Wall Street Journal, Jassy informed Treasury Secretary Scott Bessent that Amazon researchers had exploited the system to gather data usable in cyberattacks. Consequently, authorities restricted worldwide access to both the Fable 5 and Mythos 5 versions on Friday. While an Amazon spokesperson declined to disclose specific discussion details, noting that government consultations on security are routine, the move follows reports from The Information and Reuters confirming Amazon shared its concerns. David Sacks, a former AI advisor, described the situation as a direct intervention where the administration demanded Anthropic CEO Dario Amodei resolve a reported jailbreak or shut down the model, which Amodei allegedly refused to do.
This incident highlights the intensifying pressure on AI providers to align with government security directives, even when it impacts commercial operations. The involvement of a major investor like Amazon in flagging vulnerabilities suggests a shift where corporate stakeholders actively participate in regulatory enforcement. It also underscores the friction between rapid model deployment and stringent safety protocols, as Anthropic faced a choice between fixing a potential exploit or losing access to a lucrative market. The precedent set here could force other companies to prioritise government mandates over independent technical assessments, fundamentally altering the landscape of AI governance and accountability.
- Amazon CEO Andy Jassy reportedly flagged security risks in Anthropic’s Claude Fable 5 model to US officials prior to an export ban.
- Anthropic CEO Dario Amodei allegedly refused to de-deploy the model after being ordered by the administration to fix a reported jailbreak.
- The event marks a significant escalation in government oversight of AI models, with major corporate investors directly influencing regulatory outcomes.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
