The Trump administration has not lifted export controls on Anthropic's latest AI model, Claude Fable 5. The administration believes the model's guardrails can be bypassed to access powerful cybersecurity capabilities, while Anthropic argues the concerns are overblown, leaving the two sides at odds.
The Trump administration has not lifted export controls on Anthropic's latest AI model, Claude Fable 5. The controls were imposed last week over jailbreaking concerns, and high-level talks on Monday failed to resolve the disagreement.
The administration believes the model's guardrails can be disabled, allowing access to the more powerful cybersecurity capabilities of the Mythos model. The NSA reviewed the vulnerabilities and confirmed that guardrails could be removed. Amazon CEO Andy Jassy directly alerted Treasury Secretary Scott Bessent about the vulnerabilities, alarming the administration. Anthropic maintains the concerns are overblown, with co-founder Tom Brown leading discussions.
The standoff highlights tensions between the government and AI companies over frontier model safety and regulation. Anthropic is already in a prolonged dispute with the Pentagon over military use of its AI, adding political pressure. The Commerce Department is open to restoring consumer access to Fable 5, but likely only if jailbreak concerns are fully resolved.
The commenter questions why Amazon reported jailbreak concerns to the government instead of Anthropic, arguing that this is causing massive losses for Anthropic. They also suggest GPT-5.5 may be equally capable as Fable 5 and criticize Glasswing findings as overblown. While some see Amazon's actions as motivated by national security concerns, the commenter remains skeptical.