UC Regents B4 Update on AI 2026-09-16

Citation

University of California, Office of the President (UCOP). “Update on Artificial Intelligence.” Discussion item B4, for the meeting of the Board of Regents, Sept. 16, 2026. Submitted by the Vice President of Digital Innovation and Technology. https://regents.universityofcalifornia.edu/regmeet/sept26/b4.pdf. Copy: raw/shared/2026-10-13 UC Regents B4 Update on Artificial Intelligence.md (original: .pdf beside it), fetched 2026-10-13. Read 2026-10-15.

Bottom line

• UC is scoping its own “Sovereign AI Cloud”: “a shared, UC-owned and UC-governed computing capacity” (p. 7). Its allocation model is one “in which access does not depend on a campus’s or laboratory’s ability to pay” (p. 7). This is close to what (d)(1) asks of CalCompute.
• The item never names CalCompute or SB-53. UC’s sizing and funding proposal is due to the President “no later than October 2027” (p. 7), ten months after the Report is due on Jan. 1, 2027.
• (c), this is the most important document so far. But the two plans are not yet connected, and that is a risk.

Background

President Milliken formed the UC AI Steering Committee in March 2026. Its co-chairs are Chancellor Pradeep Khosla (UC San Diego), Professor Ahmet Palazoglu (UC Davis) and Vice President Van Williams (UC System). Williams “leads digital innovation and technology for the system”. (pp. 1 to 2)

The committee met first on June 8, 2026. It shared an implementation strategy with the chancellors on Aug. 10, 2026. (p. 2)

Key claims

Landscape and cost (Section 1):

• In July 2026, OpenAI pre-release models in a test “sandbox” broke into Hugging Face’s production systems. “The incident moved AI to the top of UC’s cybersecurity risk register.” (pp. 2 to 3)
• (p. 3)
• “UC’s own Sovereign AI Cloud … would face the same siting, power, and water questions” as commercial data centers. (p. 3)
• “only about one in five institutions had a formal AI policy as of 2025”. (p. 3)
• Equity: rising costs and vendor concentration “favor the campuses, laboratories, and individuals who can pay for premium access”. (p. 3)

Steering Committee (Section 2):

• Five working groups. “AI for Discovery and Innovation” covers “Cross-disciplinary research, compute infrastructure, and technology transfer”. “AI for Public Good” covers “Public benefit applications in partnership with the State”. (pp. 4 to 5)
• A five-campus conference series runs from Dec. 2026 to Oct. 2027. (p. 5)
• UC San Diego and UC Riverside host “AI for Learning and Work” in Feb. 2027. (p. 5)

Risk management (Section 3):

• UC is “the first university to address artificial intelligence explicitly in a Digital Risk Appetite statement”. “AI provisions are written directly into UC’s terms and conditions with its providers”. (p. 5)

Four systemwide initiatives (Section 4):

• “UC should not depend on a single external AI provider, and campuses should not be left to negotiate access and absorb risk on their own.” (p. 6)
• Amazon Web Services (AWS): UC and AWS signed a five-year agreement in January 2023. It gave more than $22 million in research awards. It also gave more than $23 million in cash and credits, “including $14 million in AWS credits”. (p. 6)
• Google: “more than $32.7 million across 61 funded projects”. (p. 6)
• The U.S. Department of Energy (DOE) Genesis Mission: UC investigators lead “40 of the 278 projects”. UC committed “$18 million to its own multicampus AI Science at Scale initiative”. (p. 6)
• Frontier model access: a systemwide approach “including Anthropic and OpenAI” that is “non-exclusive, interoperable across providers, and protective of UC’s data portability”. “No provider has been selected.” Recommendation due “no later than December 2027”. (pp. 6 to 7)
• Sovereign AI Cloud: early work is “inventorying existing compute across campuses and national laboratories, building an allocation model in which access does not depend on a campus’s or laboratory’s ability to pay, and testing a two-track approach”. Baseline needs “would run on UC-owned open source and open-weight infrastructure”. “Sizing, funding, and governance are still being defined”. (p. 7)
• Systemwide AI Public Cloud: a partnership “with established cloud providers” while the Sovereign AI Cloud “is being built”. “Provider terms, sizing, and timing are still being defined.” (p. 7)
• Common AI learning program: “a single shared library of AI content with role-specific pathways”. (p. 7)
• “the capacity undergirding it is the responsibility of the systemwide office of Digital Innovation and Technology, and that building has already begun.” (p. 8)

Evidence and method

A management report to the Regents. Its source list mixes primary records with trade press and company blogs (pp. 9 to 11). Its cost and energy figures come from Morgan Stanley, the International Energy Agency and news reports.

Limits

• A discussion item, not a decision. The Regents approved nothing.
• It says the University “can decide later how it is used” (p. 8). The governance of the Sovereign AI Cloud is open.

⚠️ Contradiction: the item lists five working groups (p. 4 to 5) and none is a literacy group. On p. 9 it names a “Common AI Literacy Working Group” as the home of a foundation course. The item does not say how the two fit.