Now Playing
Anthropic's Export Standoff Deepens: Talks Collapse, Legal Challenge Looms
Audio is available on Spreaker — see link below.
Anthropic's flagship models remain offline after White House negotiations collapse, raising legal questions about whether Cold War export law can apply to cloud-based AI APIs. Plus: Microsoft's China contradiction, Meta's 1.6-gigawatt bet, and DeepMind's rogue agent framework.
Audio is available on Spreaker — see link below.
Anthropic's two most advanced models are still offline, and the talks meant to fix that collapsed Tuesday with nothing resolved. That's the lead today.
The mechanism that triggered this matters too. Amazon engineers stress-tested Fable five and found a narrow potential jailbreak, something that could theoretically unlock dangerous cybersecurity capabilities.
Legal experts are now questioning whether the government had the statutory authority to issue this order at all. The core issue is whether a "deemed export" standard, a Cold War framework designed for physical technology transfers, can legally apply to a cloud-based API that anyone can access.
There's a political dimension underneath all of this. The Trump administration has been frustrated with Anthropic's public advocacy for AI regulation.
Away from the Anthropic standoff, Microsoft is quietly in a complicated position of its own. The company co-founded the Frontier Model Forum, partly to counter Chinese AI practices including model distillation.
Meta, meanwhile, signed agreements with Crusoe for one point six gigawatts of AI computing capacity across Texas and Missouri. Computing capacity measured in gigawatts is now the primary currency in this race.
One more development worth tracking. Google DeepMind released a thirty-five page framework for treating AI agents as potential insider threats, essentially applying enterprise cybersecurity logic to AI behavior monitoring.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.