Mythos is just too powerful? OK, we’ll ban it then

Anthropic’s been bigging up its new Mythos and Fable models. Mythos can hack anything, you know! So can Fable! They’re just too powerful and dangerous! So our KILLER ROBOTS can definitely answer your office email. Earlier this month, one large organisation took Anthropic’s warnings about...

Anthropic’s been bigging up its new Mythos and Fable models. Mythos can hack anything, you know! So can Fable! They’re just too powerful and dangerous! So our KILLER ROBOTS can definitely answer your office email. Earlier this month, one large organisation took Anthropic’s warnings about Mythos seriously. That organisation? The US Government. At 5:21pm, right at the end of the day on Friday 12 June, the White House directed Anthropic to block access to Mythos and Fable for anyone who wasn’t a US national. Anthropic couldn’t work out how to do this without just blocking Mythos entirely: [Semafor] According to Anthropic, the government claims to have found a “jailbreak,” or a way that users could bypass some of Fable 5’s guardrails. The White House thought Mythos was too dangerous a hacking tool to let loose on the world. Because Anthropic told everyone it was precisely that. Who told the White House there was a jailbreak? [Information, archive] Amazon CEO Andy Jassy was among the tech leaders who raised concerns to senior Trump administration officials this week about security risks in Anthropic’s most advanced models. Before this, the White House had told Anthropic to limit Mythos access for SK Telecom in Korea, claiming SK Telecom had links to China. [Wired] Either that, or the White House is still annoyed about Anthropic defying the Department of Defense in March. The government slightly loosened the restrictions on Mythos last week. Anthropic can now allow Mythos access for 100 or so US companies and agencies. But the broader restriction is still in place. [Semafor] The White House really wants control of new chatbot models. OpenAI’s new GPT 5.6 was only released to a government approved list: [blog post] Working in coordination with the government, we are starting with a limited group of trusted partners and organizations whose participation has been shared with the government before releasing the models more broadly. There’s a few things going on here: These magical new AI models are not some sort of new danger. That’s pure marketing. The US government now believes the new models are dangerous. So well done to Dario Amodei on shooting himself in the foot. The present US administration loves the AI and would love to have the magical wizard toy all to itself. The present administration is better at yelling than it is at achieving. This is yelling to prove they have the power. AI regulation’s not that hard! At least in the sense of stopping it dead. Video — Podcast

Source: Pivot to AI — Published June 29, 2026 at 20:56 UTC — Category: Business