Can AI Be Regulated?

Foreign Policy | Cameron Abadi, Adam Tooze

The U.S. federal government faces intensifying public pressure to regulate artificial intelligence, exposing a deep rift within Silicon Valley over the technology's catastrophic risks. While closed-model developers OpenAI and Anthropic advocate for federal oversight to pace the frontier, open-source proponents like Meta and Nvidia reject new strictures. This division reflects divergent commercial strategies, where closed-system advocates seek to mitigate liability and secure national champion status, while hardware providers aim to maximize compute demand.

Currently, AI spending by hyper-scalers runs between $150 billion and $200 billion per quarter, driving 36 percent of all U.S. business investment. A market reset remains a distinct possibility. Retrospective financial compensation under standard negligence law cannot address extinction-level events or systemic financial instability. Pockets of technical proficiency exist within DARPA and the National Security Agency, but keeping pace with rapid machine learning developments remains a daunting challenge for federal regulators.

Comment

The lobbying efforts by OpenAI and Anthropic for closed-model safety standards expose a fundamental tension between state-directed algorithmic containment and market-driven technological supremacy. By advocating for closed-model restrictions, these entities seek to establish a regulatory barrier that protects their capital-intensive research from open-source competitors like Meta's Llama models. This approach risks centralising technological development within a small oligopoly, potentially limiting the rapid, decentralised innovation that DARPA relies on for its software-defined military systems.

This dynamic parallels the 1990s 'Crypto Wars', where federal attempts to restrict the export of PGP encryption ultimately failed to halt the global proliferation of open-source cryptographic protocols. Similarly, imposing federal compliance burdens on American frontier labs may simply accelerate the global adoption of unrestricted alternatives, such as the open-source models developed by Alibaba's AI division.

Strategic Question for Discussion

If federal regulations restrict closed-model development in the United States, how would DARPA maintain its technological edge in software-defined military systems when open-source alternatives remain globally accessible?

The trajectory indicates that DARPA would likely bypass domestic commercial restrictions by leveraging custom, on-premise open-source architectures tailored for classified environments. However, a highly regulated domestic market could slow the pipeline of commercial talent and dual-use software innovations flowing into defence acquisition channels. Consequently, the agency might find itself increasingly reliant on legacy systems while foreign adversaries rapidly integrate unrestricted commercial models.

Share your assessment in the comments below.

To read in details click here