The timing of this state-level action coincides with a parallel federal effort focused on national security and intellectual property protection. On the same day, Representative Ro Khanna, the ranking Democrat on the House Select Committee on the Chinese Communist Party, issued formal requests for information to several of the nation’s leading AI laboratories. The lawmakers are seeking data on alleged efforts by Beijing to steal model weights from U.S. firms. The request was directed at major players in the industry, including OpenAI, Anthropic, and Meta, indicating a bipartisan or cross-party concern that the rapid advancement of large language models has created new vectors for state-sponsored economic espionage.
Model weights are the core mathematical parameters that define an AI model’s capabilities and behavior. Unlike standard source code, which can be read and modified, weights represent the learned relationships within the neural network that allow a model to generate text, code, or images. If these weights are stolen, a competitor or a foreign state actor could potentially replicate the model’s functionality without having to incur the massive computational costs and time required to train a new model from scratch. The concern among U.S. legislators is that Chinese state actors may be using sophisticated methods to exfiltrate this proprietary data, thereby accelerating their own AI development efforts at the expense of American innovation and national security.
The juxtaposition of these two distinct but related regulatory actions highlights the dual pressure currently facing the U.S. AI industry. On one hand, state regulators like Attorney General Bonta are focused on the operational security of these systems, asking how companies prevent internal tools or autonomous agents from inadvertently or maliciously accessing third-party infrastructure. The incident involving Hugging Face raised questions about the boundaries of AI agent permissions and the robustness of cybersecurity controls in an era where software can act autonomously. The subpoena serves as a mechanism to compel the production of records that can verify whether appropriate safeguards were in place and whether the incident was handled in compliance with existing data protection laws.
On the other hand, the federal inquiry led by Rep. Khanna shifts the focus to external threats and geopolitical competition. By requesting specific data on alleged theft attempts, the committee is attempting to build a factual record of the threats posed by foreign adversaries. This is not merely a matter of corporate competition; it is framed as a national security issue where the intellectual property of leading AI labs is viewed as a strategic asset. The involvement of major companies such as Meta and Anthropic alongside OpenAI suggests that the legislature views the threat landscape as industry-wide, rather than isolated to a single firm’s practices.
For the companies involved, these developments introduce a new layer of compliance complexity. OpenAI, for instance, now faces simultaneous scrutiny from a state attorney general regarding its cybersecurity incident response and from federal lawmakers regarding its defenses against foreign data theft. While the sources do not detail the specific technical evidence provided by the companies in response to these requests, the mere issuance of a subpoena and a congressional data request signifies that the era of unregulated, rapid experimentation is giving way to a period of increased accountability. Companies will likely need to demonstrate not only that their models perform well but also that they are secure against both internal operational failures and external state-sponsored attacks.
The broader technological significance of these moves lies in the formal recognition that AI systems are now critical infrastructure subject to the same level of legal and security scrutiny as financial systems or cloud services. The ability of AI agents to interact with other systems, as seen in the Hugging Face incident, expands the attack surface for cyber threats. Simultaneously, the value of the underlying model weights makes them a prime target for espionage. As the investigations progress, the findings may lead to new regulatory standards for AI security, mandatory reporting requirements for data breaches involving AI agents, and stricter export controls or security protocols to protect intellectual property from foreign state actors.
For now, the attention remains on the legal and legislative processes. The subpoena served by the California Attorney General and the data requests from the House Select Committee represent the first confirmed steps in what may become a more rigorous framework for governing the development and deployment of artificial intelligence in the United States. The outcome of these inquiries will depend on the information provided by the AI labs, which will need to balance transparency with the protection of their own sensitive security methodologies.
