Amazon’s internal security research set off a chain reaction that landed Anthropic’s flagship AI models, Fable 5 and Mythos 5, in the crosshairs of the White House. A paper released by the e‑commerce giant showed that, through a series of carefully crafted prompts, the models could produce information that might aid cyber attackers. The findings were passed to senior officials by Amazon chief Andy Jassy, prompting an export‑control directive that bars foreign nationals from accessing the two systems.
The move immediately affected Anthropic’s workforce. A sizable share of the company’s researchers were born abroad, and the new rule effectively cut them off from their own products. In response, Anthropic rejected the government’s characterization of the issue as a “jailbreak.” The company argues that similar weaknesses could be uncovered in any large language model, including publicly available versions such as GPT‑5.5. Security veteran Katie Moussouris echoed that sentiment, posting on BlueSky that she had reviewed the Amazon paper and concluded, “It’s not a jailbreak.”
Former Commerce Department official Kate Koren suggested that the White House’s decision may have been influenced by a broader dislike of Anthropic. The friction is not new. Earlier this year, the Trump administration ordered federal agencies to stop using Anthropic’s AI, citing concerns over mass surveillance and lethal autonomous weapons. Hours later, Defense Secretary Pete Hegseth labeled the firm a supply‑chain risk. Those actions marked a stark departure from a period of cooperation during which the government and Anthropic worked together to expand access to the Mythos model.
Now the two sides appear poised for another showdown. While Anthropic points to the prevalence of similar vulnerabilities across the AI landscape, the administration is focused on the potential national‑security ramifications of unchecked model capabilities. The episode underscores the growing tension between rapid AI innovation and the government’s effort to impose safeguards on emerging technologies.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.