The Special Competitive Studies Project (SCSP) is a 501(c)(3) nonprofit national-security think tank founded in 2021 by Eric Schmidt as the nongovernmental successor to the National Security Commission on Artificial Intelligence (NSCAI). Headquartered in Arlington, Virginia, it advances a China-focused agenda on AI and emerging-technology competition and is known for its Mid-Decade Challenges reports and its framing of US AI policy as a race with China.
| Field | Value |
|---|---|
| Type | 501(c)(3) nonprofit national-security think tank |
| Headquarters | Arlington, Virginia |
| Founded | 2021 |
| Founder | Eric Schmidt |
| CEO | Ylli Bajraktari (former executive director of NSCAI) |
| Known for | Mid-Decade Challenges reports; the "race with China" framing of US AI policy; technology-competition agenda |
Origins
SCSP was founded in 2021 by Eric Schmidt and senior staff from the National Security Commission on Artificial Intelligence (NSCAI), which had concluded its congressionally chartered work in 2021 after delivering its final report. SCSP serves as the nongovernmental successor vehicle, carrying over the same agenda and personnel and continued funding by Schmidt. SCSP describes its mission as making "recommendations to strengthen America's long-term competitiveness for a future where artificial intelligence (AI) and other emerging technologies reshape national security, the economy, and society." Its core frame is strategic competition with China in AI, semiconductors, biotech, and adjacent domains.
Activities and positions
SCSP's principal publications include Mid-Decade Challenges to National Competitiveness, a flagship annual or biennial report in the NSCAI tradition; the Global AI Race Tracker, an ongoing project tracking US-China capability, investment, and talent metrics; and a series of Action Plans recommending federal legislative and executive actions across compute, talent, data, and export controls.
On policy, SCSP treats compute as the primary lever, a position aligned with the Center for Security and Emerging Technology (CSET), Center for Strategic and International Studies (CSIS), and RAND Corporation compute-governance school but framed more explicitly around competition. It supports aggressive export controls, including tightening BIS controls and closing loopholes. On industrial policy, it argues for federal investment in domestic chip manufacturing, data center power infrastructure, and AI deployment in government. Among national-security commentators, SCSP is often among the most assertive voices on the speed and magnitude of the China challenge.
Relative to adjacent institutions, SCSP is more overtly competition-framed than Center for Strategic and International Studies (CSIS) or Brookings Institution AI Initiative, more policy-advocacy oriented than RAND Corporation (which is primarily research-service), and more Schmidt-aligned than Center for Security and Emerging Technology (CSET) (which is foundation-funded but Georgetown-housed). Its staff overlap with the Department of Defense, including Chief Digital and Artificial Intelligence Office (CDAO) leadership lineages.
Agentic AI governance assessment (2026)
In a June 18, 2026 assessment, "Agentic AI: The Governance Race We Must Not Lose", SCSP president Ylli Bajraktari and Rama Elluru argued that the United States is governing agentic AI too slowly relative to China and that the gap is widening . The assessment frames the competitive threat less as China fielding agentic systems first than as adversaries deploying them in the domains where U.S. governance is weakest — cyber operations, influence campaigns, and economic-intelligence collection, where feedback loops are tight and capabilities improve fastest. It contrasts China's reliance on existing regulation and standards rather than agentic-specific hard rules, which SCSP reads as preserving deployment flexibility.
The assessment urges policymakers to govern "agentic capability" rather than the underlying model, on the grounds that risk lives in the scaffolding built around a model — connectors to real-world systems, persistent memory, planning, permission structures, and guardrails — so that two systems running the same model can behave very differently. It identifies an "accountability deficit" across three fronts: responsibility is untraceable across chains of agents and sub-agents, evaluations measure task completion rather than safety or manipulability, and privacy harms are structural because agents accumulate inferential profiles of the people they serve. Its recommendations include fully funding and codifying the Center for AI Standards and Innovation (CAISI) agentic-evaluation mandate, using federal procurement (OMB and GSA) to require tamper-evident action logs in agentic systems sold to the government, establishing accountability architecture that maps each deployed agent to an identifiable liable legal person, closing the AI-talent gap across sectoral regulators and national-security agencies, and coordinating standards with allies. On the accountability point it cites the Air Canada tribunal ruling, which held the airline liable for its chatbot's misinformation, as a signal that institutions will not be able to disclaim responsibility for their agents "even though the legal and technical architecture to operationalize that does not yet exist," and models its proposed non-delegable liability on financial-services practice.
The assessment also argues that governance and speed are complementary rather than opposed: a cascade of high-profile failures, or one catastrophic failure, "could produce a public and political reaction that sets back beneficial applications for years," citing the roughly six-year stall in self-driving vehicles after the 2018 Tempe, Arizona fatality, so that "the absence of serious governance creates the conditions for the kind of failure that produces overcorrection." On trajectory it describes capability advancing as "a heat map, not a tide" — fastest where answers can be verified — expecting near-term movement into paperwork-intensive professional work and agent orchestration, and medium-term entry into robotics and healthcare use cases existing frameworks were not designed for. This was the basis for press coverage on June 22, 2026 reporting SCSP's warning that lagging action on agentic-AI governance threatens U.S. competitiveness (Source: insideaipolicy.com).
People
Eric Schmidt is SCSP's founder and chair; the former Google CEO also funds Schmidt Futures, Schmidt Sciences, and the Horizon Institute for Public Service. Ylli Bajraktari, the former NSCAI executive director, is CEO. Robert Work, a former Deputy Secretary of Defense and longtime proponent of AI and autonomy in defense, serves as a senior advisor. Nadia Schadlow and H.R. McMaster are among the affiliated national-security experts. Staffing draws heavily on former NSCAI, DoD, and intelligence-community personnel.
Funding
SCSP is funded primarily by Eric Schmidt and the Schmidt philanthropic ecosystem. Additional corporate and foundation support is listed in SCSP disclosures.
Reception
Critics argue that SCSP's concentration of funding and its alignment with Schmidt's views and business interests warrant scrutiny, particularly given Schmidt's investments in defense AI and emerging tech. Critics also argue that the China-race framing drives policy toward confrontation rather than coordination. Defenders point to bipartisan staffing and to alignment with the broad consensus of the US-government national-security community on the China challenge.
Relationships
- related: Horizon Institute for Public Service — parallel Schmidt-ecosystem investment in tech policy talent
- related: Center for Strategic and International Studies (CSIS), Center for Security and Emerging Technology (CSET), RAND Corporation — adjacent natsec-AI shops
- related: Chief Digital and Artificial Intelligence Office (CDAO), US AI Safety Institute (NIST AISI) — federal counterparts SCSP advises
- related: America's AI Action Plan, Executive Order 14365 — Ensuring a National Policy Framework for AI — federal policies shaped partly by SCSP-adjacent advocacy
- related: Export Controls (AI), AI Race Dynamics, AI and National Security, Semiconductor Supply Chain
- related: NSCAI (predecessor; not a current entity)
Notes
This page draws on SCSP's published policy footprint plus its June 18, 2026 agentic-AI governance assessment, cited inline. The assessment is queued for foundational ingest; no SCSP report has yet been folded in through the formal ingest path, so confidence remains medium pending that step.