The Regulatory Dilemma of Open-Source AI: Anthropic CEO's Congressional Testimony
On June 28, Anthropic co-founder and CEO Dario Amodei issued a stark warning to U.S. lawmakers: open-source AI development is sliding down a 'very dangerous path.' During a congressional hearing, he argued that once a sufficiently capable AI model is released under an open-source license, the original developers effectively lose all meaningful control over its usage. 'We cannot track who is using the model, for what purpose, and we certainly cannot revoke access unilaterally or push security updates after a vulnerability is discovered,' Amodei stated. He emphasized that while such risks are manageable in closed-model ecosystems, fully open models create a governance vacuum that can lead to irreversible abuse.
Why Crypto Is Front and Center: The Governance Void in On-Chain AI
Amodei's testimony resonated strongly within the crypto community. The convergence of decentralized finance (DeFi) and artificial intelligence is accelerating: on-chain AI agents execute trading strategies automatically, decentralized compute markets support open-source model training, and zero-knowledge machine learning (ZKML) verifies off-chain inference results. All these applications depend on the openness and composability of open-source models. However, if these models lack built-in backdoors for emergency intervention or update channels, once deployed into smart contracts, no subsequent security patch can be applied. For example, a maliciously tampered open-source trading model could run perpetually on-chain, causing asset losses with no way to halt it. In contrast, closed models—such as Anthropic's Claude series—maintain continuous monitoring through API layers and dynamically adjust behavioral thresholds, making them more suitable for capital-sensitive crypto applications.
Open-Source vs. Closed Models: A Risk Trade-off
Amodei explicitly contrasted the two paradigms: closed models allow developers to sustain 'continuous control,' including real-time anomaly detection, automatic blocking based on usage signals, and vulnerability fixes via model updates. Open-source models forfeit these mechanisms upon distribution. While open-source promotes technological democratization and audit transparency, in financial core scenarios—such as asset custody, oracle price feeds, and automated market making—irrevocable open-source licenses can become systemic risks. Crypto projects selecting underlying AI models must incorporate 'governance sustainability' as a core decision metric: does the model allow revocation capability? Is the update mechanism compatible with on-chain governance? Can model weights be urgently delisted? The answers will define the security baseline of the emerging DeFAI (Decentralized Finance + AI) sector.
As of press time, the U.S. Congress has not reached consensus on a new AI regulation bill, but Amodei's remarks have pushed lawmakers to pay closer attention to potential abuse scenarios of open-source models in financial infrastructure. The crypto industry should proactively establish 'governable AI' standards to avoid becoming a testing ground for uncontrollable open-source deployment.