Key facts
- The White House imposed export controls on Anthropic's new AI model, Fable.
- Concerns about the model's security guardrails were raised by Amazon.
- Anthropic CEO Dario Amodei defended the model's safety features during calls with administration officials.
- The administration cited national security authority for the export control order.
- Anthropic stated the order resulted in the disabling of its Fable and Mythos 5 models for all customers.
The White House imposed export controls on Anthropic's newly released AI model, Fable, following a rapid series of high-level discussions and concerns over its security guardrails. The decision came after Amazon CEO Andy Jassy reportedly alerted the administration to potential vulnerabilities in the model's defenses.
Over a frantic 24-hour period, senior administration officials, including Treasury Secretary Scott Bessent and White House Cyber Director Sean Cairncross, engaged in tense calls with Anthropic CEO Dario Amodei. Amodei defended the model's safety, arguing that bypasses were specific and not a broad 'jailbreak,' and that the system's safeguards were robust. However, officials, backed by findings reviewed by the National Security Agency, expressed doubts about the security of the AI's guardrails.
Despite Amodei's arguments and Anthropic's public statements defending its systems, the administration viewed the situation as a national security risk. Officials described the export controls as a last resort after attempts to persuade Anthropic to voluntarily pull the model failed. The company stated that the order effectively disabled its Fable and Mythos 5 models for all customers to ensure compliance, calling the action disproportionate and lacking transparency.
Anthropic has positioned itself as an advocate for AI regulation, but in this instance, the administration felt the company was not taking potential risks seriously enough. The White House emphasized that while innovation is a priority, security must also be paramount.