Anthropic has discontinued its Fable and Mythos AI model lines following a directive from the Trump administration. The Commerce Department cited national security concerns, specifically citing a potential "jailbreak" vulnerability in Fable 5 that officials worried could be exploited.
The decision reflects escalating government scrutiny of advanced AI systems. Federal officials flagged that the jailbreak in Fable 5, which allows users to circumvent built-in safety guardrails, posed risks if the model fell into adversarial hands or was used for harmful purposes. Rather than pursue fixes or continued development, Anthropic chose to retire both model families entirely.
Fable and Mythos represented specific branches of Anthropic's research and product portfolio. The shutdown removes them from public access and future development roadmaps. This marks a tangible example of regulatory intervention shaping which AI systems companies deploy, moving beyond voluntary guidelines into direct government action.
The Commerce Department has intensified its oversight of AI development since the Trump administration took office. This move signals that federal agencies now view certain AI capabilities as falling under national security jurisdiction, similar to semiconductors or cryptography. The calculus for AI companies is shifting: maintaining a model line may trigger regulatory friction that outweighs commercial benefit.
Anthropic's response to retire rather than fight or remediate suggests the company assessed the compliance burden as prohibitive. The decision also raises questions about how broadly the administration will apply similar pressures. If jailbreak vulnerabilities trigger shutdowns, many models across the industry could face similar scrutiny.
The timing places AI safety enforcement directly under national security authorities rather than leaving it to industry self-regulation or traditional product liability frameworks. This centralizes decision-making power but also creates uncertainty for developers about which vulnerabilities trigger action.
For users relying on these models, the shutdown forces migration to alternative systems. For Anthropic,