For makers and artists relying on frontier models to generate code, design assets, or research, the current political standoff between Washington and Anthropic introduces a dangerous bottleneck. If the US government insists that large language models be proven “unhackable” before deployment, it risks stalling innovation while leaving creators vulnerable to the very prompt injection attacks that security experts say are unfixable. The demand for absolute security ignores the reality that no AI system is immune to manipulation, yet officials are treating the release of Anthropic’s Fable 5 as a breach of trust rather than a standard product launch.
A conflict over timing and oversight
Reports emerging from Axios suggest that administration officials are accusing Anthropic of ignoring a recent cyber executive order from the Trump administration. This directive called for voluntary government oversight of AI models, with a designated clearinghouse intended to sign off on releases. While Anthropic welcomed the concept of oversight, they proceeded with the launch of Fable 5 before the clearinghouse was operational. Government sources now claim this sequence of events demonstrates a disregard for safety protocols.
One official summarised the tension with stark words: “Everybody said Anthropic was a bad actor. Some of us said it was time to give them a chance. Now those people are questioning that. They screwed us.”
The administration also alleges that Anthropic knew a “jailbreak”-a method to bypass safety filters-could occur. According to the narrative, “They came to every fork in the road and took the wrong fork.” This warning reportedly originated from Amazon and other technology firms. However, neither the existence nor the severity of this specific jailbreak has been independently confirmed.
Communication between the two sides has been described as fractured. A government source told Axios, “It’s like they just speak in different languages.” Talks are currently underway between the Department of Commerce and Anthropic staff, with further meetings involving the CIA and science advisor Michael Kratsios scheduled.
The myth of unhackable models
The core accusation-that Anthropic knew of the jailbreak risk and remained silent-reveals more about the government’s understanding of artificial intelligence than it does about Anthropic’s conduct. Anyone working closely with these systems understands that they can be hacked. OpenAI has previously warned that prompt injection, a related hacking technique, may never be fully solved. There is currently no fix for LLM security that guarantees immunity.
The critical question remains how severe the breach is and how quickly countermeasures can be deployed. If the US government insists that frontier AI models must be “unhackable” before they are shipped internationally, difficult negotiations lie ahead. This stance ignores a warning issued by Anthropic CEO Dario Amodei in 2023, who stated that “a jailbreak could be life or death” if someone managed to bypass safety protocols in science, technology, and biology.
Industry pushback
More than 100 cybersecurity experts and technology industry executives have published an open letter to Trade Secretary Lutnick and National Cyber Director Cairncross. They are calling for the immediate lifting of export controls on Fable and Mythos. The signatories argue that while Anthropic’s models are proficient at identifying security flaws in software, they are not uniquely superior at it. Competing models, including GPT-5.5, Opus, Sonnet, and the Chinese Kimi 2.7, possess similar capabilities.
Furthermore, Anthropic implemented several safeguards into Fable that the security community dismissed as overkill on the day of launch. The letter warns that export controls are stripping defenders of the best tools while Chinese open-weight models remain only months behind the top US offerings. Notable signatories include Alex Stamos of Corridor, Rachel Tobac of SocialProof Security, Katie Moussouris of Luta Security, Dan Lorenc of Chainguard, and Joe Levy of Sophos.
Key takeaways
Government officials are accusing Anthropic of ignoring a cyber executive order by releasing Fable 5 before the designated oversight clearinghouse was established.
Claims that Anthropic knew about a jailbreak and stayed silent highlight a fundamental misunderstanding of AI security, as prompt injection vulnerabilities are currently considered unfixable.
Over 100 security experts and executives are urging the US to lift export controls on Anthropic models, arguing that restricting access removes vital tools from defenders while Chinese alternatives catch up rapidly.
Anthropic CEO Dario Amodei previously warned that bypassing safety protocols in scientific and biological contexts could have life-or-death consequences, making the demand for “unhackable” models impractical.
