House subcommittee hearing flags DeepSeek as wake-up call on Chinese AI, urges investment in research, export enforcement and data protections

The House Research and Technology Subcommittee heard bipartisan testimony Tuesday that the rapid rise of the Chinese AI company DeepSeek represents a strategic challenge to U.S. technological leadership and underscores gaps in export controls, federal research capacity and data-governance safeguards.

Experts called DeepSeek as a "Sputnik moment" for artificial intelligence, while urging a mix of accelerated public investment and improved regulatory and enforcement tools so the United States can both compete and reduce security risks.

Why it matters: witnesses said DeepSeek's open technical disclosures and low-cost performance make it likely to be widely adopted internationally, raising concerns about user data stored in China and the potential for agent-like models to introduce undetectable vulnerabilities into software and critical systems.

Adam Thier, a senior fellow at the R Street Institute, told the panel that DeepSeek's January launch "sent shockwaves for tech markets and policy circles alike," and argued the U.S. should prioritize an "opportunity-oriented" agenda to keep freedom to innovate while avoiding state and local regulatory fragmentation.

Gregory Allen, director of the Wadewani AI Center at the Center for Strategic and International Studies, emphasized export controls as a key lever. He said U.S. actions to limit sales of advanced AI chips to China have had effect but were sometimes poorly timed and under-resourced, allowing Chinese firms to amass stocks of components or to rely on slightly downgraded chips designed to evade controls.

Julia Stojanovich, director of the Center for Responsible AI at New York University, focused on data governance and transparency, saying that model weights alone are insufficient for assessing risk: "Releasing model weights alone is not enough," she testified. She warned that DeepSeek stores user prompts, device information and IP addresses on servers in China and offers limited public documentation about retention or reuse, creating a data-sovereignty vulnerability for U.S. citizens and enterprises.

Tim Fist of the Institute for Progress warned about the coming era of agentic AI and the possibility of backdoors that are currently difficult or impossible to detect. He urged stepped-up federal R&D on interpretability and secure hardware, and creation of a rapid analysis capability within a trusted agency such as NIST.

Repeated themes

- Export controls and enforcement: Witnesses said controls on advanced chips remain the strongest near-term tool to limit adversarial hardware access, but they must be implemented with more technical agility, better interagency capacity and multilateral coordination. Gregory Allen described three vital elements for effective export controls: "authority, capacity, and will."

- Public R&D and compute resources: Panelists urged full funding and rapid scale-up of the National AI Research Resource (NAR R), National Science Foundation AI investments and other federal programs to provide researchers and small companies with data, compute and training to compete with lower-cost foreign systems.

- Data governance and transparency: Experts called for stronger disclosure and measurement standards so model behavior, training data provenance and data-retention practices can be audited. Julia Stojanovich said, "Without transparency, we can't check whether data is being handled in accordance with legal and ethical standards."

- Workforce and talent: Several witnesses and members stressed the need to attract and retain international STEM talent and to expand AI literacy and education domestically to sustain a skilled pipeline.

What lawmakers discussed: Members pressed witnesses on whether DeepSeek used U.S. chips and on the veracity of cost estimates publicized for training models. Witnesses cautioned that claimed training costs (widely reported at about $5.6 million) are likely cherry-picked and do not reflect the full R&D and experimentation expenses behind a production system. Committee members also raised concerns about recent federal staffing and research-funding changes at agencies such as NIST and NSF and asked whether those actions weaken U.S. competitiveness.

No formal actions were taken at the hearing. Members and witnesses offered policy options discussed during testimony including strengthening NIST's measurement role, completing and funding the National AI Research Resource at scale, sharpening and resourcing export-control enforcement, and preserving federal support for university research and PhD training.

Ending: Witnesses and members framed the DeepSeek episode as both a challenge and an opportunity: a prompt to accelerate U.S. public investment in compute and research, clarify data-governance rules, and coordinate export controls and technical measurement to reduce security risks without stifling innovation.