On Friday, June 12, AI giant Anthropic was instructed to roll back the release of its most advanced AI models including Fable 5 and Mythos by the Trump administration. According to a report by Axios, the Trump administration asked Anthropic for the roll back after the officials cited a ‘national security threat’. As per the Axios report, Anthropic received a call at 1 pm ET giving just 90 minutes to comply before the new licensing controls would be imposed. Anthropic had already worked with the government on pre-release testing of Fable and received explicit approval to deploy the model. However, the sudden reversal left the executives scrambling. “We immediately sought to understand the specific nature of the threat so we could remediate it,” an Anthropic source said, though officials held firm on the demand.By 5.30 pm ET, the Commerce Department issued a letter imposing restrictions on where the models could be used and by whom.
National security concerns
This move of the Trump administration underscores the government’s rapid response to warnings from Amazon and other companies about the capabilities of frontier AI systems. Officials have grown increasingly concerned about how powerful models could be misused, prompting emergency measures even after prior approvals. Following the directive, Anthropic began the process of de-deploying the models. The incident highlights the tension between AI labs pushing forward with cutting-edge releases and regulators racing to establish safeguards.
Anthropic challenges US decision
In a statement published after the Trump administration’s order to ban Mythos 5 and Fable 5, Anthropic said it disagrees with the government’s decision, adding that the concerns appear to be linked to a “jailbreak” technique that allowed researchers to identify a small number of software vulnerabilities.According to a Wall Street Journal report, the jailbreak research was carried out by researchers at Amazon. Katie Moussouris, chief executive of cybersecurity firm Luta Security, said the researchers used a series of prompts to make Anthropic’s model provide information about a handful of security vulnerabilities.Anthropic had reportedly shared a copy of the report with her, she said. The company has maintained that the vulnerabilities identified were already known and relatively minor.“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities,” Anthropic said in a statement. The company argued that similar results can be achieved using other publicly available AI models and said the vulnerabilities involved were minor and already known.Anthropic has complied with the directive but said it hopes access to the models can be restored. The company has also warned that applying similar standards across the industry could slow the release of future advanced AI systems.
