What happened
Rep. Ro Khanna, whose district covers a large chunk of Silicon Valley, threw his weight behind a federal bill that would ban the creation of recursive self-improving AI systems, according to CryptoBriefing's Monday report. Recursive self-improvement, in the technical sense the bill targets, refers to AI models capable of autonomously rewriting or retraining their own architecture to become more capable without human intervention.
It's the specific capability that AI safety researchers at labs including Anthropic and DeepMind have flagged as the pivot point between tool-grade AI and systems that outpace oversight. Khanna's office framed the endorsement as a preemptive move, not a response to a specific incident. The congressman has been vocal on tech policy for years, but this is the first time he has publicly aligned with the safety camp on a hard capability ban rather than a disclosure or licensing regime.
Why it matters
A hard ban on a specific AI capability class is a materially different policy instrument than what Washington has floated so far. The Biden-era executive order and the subsequent AI Bill of Rights leaned on transparency, red-teaming, and reporting thresholds. A statutory prohibition, if it clears both chambers, would put the US closer to the EU's AI Act in posture and further from the light-touch approach the industry has lobbied for.
For crypto, the read-across is direct. Decentralized AI protocols pitch themselves as an alternative to centralized labs, and any federal rule that constrains what any US-linked entity can train has implications for the compute, model-weight, and inference markets those tokens are built around. The bill also lands as the Trump administration's AI czar David Sacks continues to push a pro-innovation line, setting up a policy collision that markets will price.