Days after restricting its Fable 5 and Mythos 5 AI models for non-Americans, the Trump administration has reportedly issued a clear ultimatum to Anthropic: if the company wants the government to lift the strict controls placed on its most advanced AI models, it must stop arguing and start fixing the software’s security flaws. According to a report by Wired, government officials made it clear that the company must actively patch “jailbreaking” – a technique where users employ specific text prompts to trick an AI into bypassing its built-in safety rules.While Anthropic has repeatedly claimed the government’s concerns are overblown and the real-world risks are minimal, the White House appears to be no longer interested in debating. Following a high-level technical meeting on Monday (June 15) between Anthropic, the Commerce Department and National Cyber Director Sean Cairncross, officials made it clear that the ball is entirely in the tech company’s court, the report claims.
A threat to national security
The administration’s firm stance is backed by a recent investigation by the National Security Agency (NSA). The NSA officially concluded that tech-savvy users can successfully disable the guardrails on Fable 5. What’s worrying is that these guardrails were specifically designed to prevent foreign users from accessing highly sensitive capabilities inherited from Anthropic’s more powerful “Mythos” model.Without those digital fences, the government fears the AI could be weaponised to assist in dangerous fields like automated hacking, advanced chemistry, and biology. Citing sources familiar with the internal discussions, the report claims that the administration now views the entire situation as Anthropic’s problem to solve.Officials pointed out that neither the National Security Agency nor the Commerce Department’s newly formed Center for AI Standards and Innovation has the workforce or the time to hunt down every single loophole on every piece of software that hits the market. Instead, Washington expects Anthropic to be far more proactive.
Experts say White House demanding the impossible
While the White House’s demands are clear, independent cybersecurity experts reportedly warn that the government might be asking for the mathematically impossible. A growing number of computer scientists believe that current AI safety guardrails are merely temporary band-aids. They argue that as long as humans or other AI systems are smart enough, they will always find a creative way to outsmart a chatbot’s rules.
Leave a Reply