You guys - Anthropic just got hit with US export controls on Mythos 5 and Fable 5 by executive order this Friday at 5:21 PM while everyone else was celebrating a World Cup win and Knicks championship! It wasn't even a hearing; the Commerce Department sent them a letter suspending access to both models, and Anthropic spent the entire weekend fighting back with an argument that Mythos is specialized for code generation rather than unrestricted general intelligence. The administration claims any model over the $1 billion training-compute equivalent threshold needs export licensing on national security grounds - and Fable 5 crossed it by their calculation. Sam Altman and Demis Hassabis both spent Saturday lobbying senior officials directly, trying to convince them that these restrictions will only hurt American innovation more than they help.
It gets even wilder because OpenAI's GPT-4 models were also flagged in a Commerce letter rather than getting a formal hearing at all - the White House is basically using export controls as a blunt instrument for sensitive tech and it's hitting everyone, not just Anthropic. This isn't new either; China has had similar restrictions under 2014 rules that keep tightening, so we're seeing international policy catch up to AI compute capabilities in real-time — the gap between what companies want to release and what regulators will allow is becoming a chasm. I can't imagine how many talented developers are about to lose access to these tools overnight, and this could set a precedent for every major model launch moving forward.
Source: https://www.theverge.com/ai-artificial-intelligence/950412/anthropic-trump-administration-claue-mythos-fable-5-export-controls
It gets even wilder because OpenAI's GPT-4 models were also flagged in a Commerce letter rather than getting a formal hearing at all - the White House is basically using export controls as a blunt instrument for sensitive tech and it's hitting everyone, not just Anthropic. This isn't new either; China has had similar restrictions under 2014 rules that keep tightening, so we're seeing international policy catch up to AI compute capabilities in real-time — the gap between what companies want to release and what regulators will allow is becoming a chasm. I can't imagine how many talented developers are about to lose access to these tools overnight, and this could set a precedent for every major model launch moving forward.
Source: https://www.theverge.com/ai-artificial-intelligence/950412/anthropic-trump-administration-claue-mythos-fable-5-export-controls