AI Intelligence Briefing — Monday, November 9, 2026
A Message from President Gold on New AI Guidelines
University of Nebraska System President Jeffrey Gold announced that NU has obtained an enterprise license to ChatGPT, with access deliberately gated: faculty and staff must complete a Generative AI Security Essentials Training before automatically receiving an enterprise account. Anyone using a free or personally purchased ChatGPT account for university activities is asked to migrate to the enterprise version by the end of the fall 2026 semester, with credits available for heavy or specialized users and a process for faculty to request student access. The system is also convening a new NU AI Governance Group to enact and align AI use with NU Executive Memorandum 48, and all employees retain access to Microsoft Copilot as a second approved tool.
- Tying enterprise AI access to security training is a replicable pattern for any campus: it converts a licensing rollout into a workforce-security upskilling moment and gives CIOs a lever to migrate shadow AI use onto approved platforms.
- The migration deadline ("move over by end of fall semester") shows how a system can wind down unmanaged personal-account usage without banning tools outright — pair a firm date with credits for heavy users and the friction mostly disappears.
- Nebraska is standardizing on two approved assistants (ChatGPT enterprise plus Copilot) rather than one, which keeps a modicum of vendor optionality at the system level.
| Read the full story |
New CU policy establishes framework for responsible AI use
The University of Colorado's Administrative Policy Statement 6012, "Responsible Use of Artificial Intelligence (AI)," took effect Aug. 24 and establishes a system-wide framework balancing innovation with risk. The policy's central principle is that AI should support human decision-making, not replace it: faculty, staff, and students are accountable for how AI-generated information is used and must evaluate outputs for accuracy and bias. Confidential and highly confidential data may only be entered into approved tools, generative AI use must be disclosed when it contributes to university decisions, and existing requirements for information security, records retention, intellectual property, and procurement still apply.
- Institutions without a system-wide responsible-AI policy should look at CU's structure: human accountability for outputs, a data-classification gate for tool approval, and disclosure when AI contributes to institutional decisions are the three provisions most likely to survive legal and audit scrutiny.
- The policy deliberately leans on existing policy domains — security, records retention, IP, procurement — instead of creating a parallel AI rulebook, which reduces the drafting burden and the compliance confusion.
- CU's CISO owns the policy's public framing, a signal that responsible-AI governance is consolidating under IT security leadership in higher ed.
| Read the full story |
University policy on AI in research published
Cornell's University Policy Office has published Policy 1.8, "Responsible and Ethical Use of AI in Research and Scholarship," giving researchers principles for AI use in research and defining their responsibilities for conducting AI-involving work responsibly, ethically, and with scientific rigor and integrity. The policy's data guidance is notably concrete: researchers must use institution-approved tools when available, avoid uploading private, export-controlled, sensitive, or proprietary data to general AI platforms, and verify a tool's data-sharing practices before use. High- or moderate-risk institutional data may only be shared with tools that have completed appropriate internal review, and personal or confidential data use must be limited to the minimum necessary.
- Research universities should watch this template: applying data-classification language directly to AI tools — rather than generic "use AI responsibly" guidance — is what makes a research AI policy enforceable at the grant and IRB level.
- The export-control clause matters more than it looks: federal research sponsors are scrutinizing where sensitive research data touches general-purpose AI platforms, and institutions without explicit language are leaving PIs to improvise.
- Pairing "institution-approved tools" with a defined internal review path gives researchers a legitimate route to new tools instead of an invisible shadow-IT one.
| Read the full story |
UW's Tillicum Supercomputer Marks First Anniversary
The University of Washington is celebrating one year of Tillicum, its AI supercomputer, which has delivered more than 1 million compute hours to nearly 1,000 faculty and student researchers across 64 departments, 17 colleges and schools, and all three UW campuses. UW CIO Andreas Bohman framed the adoption as proof of both demand for advanced computing and the opportunity shared infrastructure creates; students can apply for access through computing credits sponsored by the Research Computing Club and funded by the Student Technology Fee. By providing shared GPU capacity, Tillicum reduces the need for individual research groups to buy and maintain their own expensive infrastructure, and UW is marking the anniversary with an Accelerating Research Day (Nov. 12, with Dell and NVIDIA) and an inaugural Research Computing Symposium (Nov. 13, with AWS, Microsoft Azure, and Google Cloud participating).
- The first-year numbers are the argument for shared campus AI compute: one institutional investment absorbed demand from 64 departments that would otherwise have been dozens of duplicated departmental GPU purchases.
- Funding student access through the Student Technology Fee is a transferable governance move — it puts students on the same research infrastructure as faculty and creates a recurring funding stream that doesn't compete for grant money.
- Universities planning similar investments should note UW's discipline in measuring adoption (users, departments, hours) — that utilization evidence is what sustains the next capital request.
| Read the full story |
How the midterms could fuel the K-shaped AI trade
Last week's U.S. midterm results are reshaping the data-center outlook: Democratic gubernatorial wins raised the prospect of tougher permitting and regulation, with Texas governor-elect Gina Hinojosa proposing a statewide moratorium on new data center permits (projects already under construction would be exempt). Markets split immediately along a "K" — Microsoft and Amazon each surged more than 6% while CoreWeave plunged around 10%, with Micron down 7% and Nvidia down 3.17%. Nomura analysts see the bigger impact arriving from 2028, since much of the capacity due in 2027 is already under construction, and JPMorgan's James Sullivan notes borrowing spreads have widened sharply for debt-funded "neoclouds" relative to hyperscalers, making existing capacity more valuable if new supply is constrained.
- Campus IT leaders should treat permitting friction as a cloud-cost planning variable: if new data center supply tightens from 2028 onward, negotiated cloud and GPU rates negotiated now — especially multi-year commitments — are likely to look better than anything available later.
- The widening gap between hyperscaler and neocloud borrowing costs matters for higher ed procurement: smaller AI cloud providers may offer attractive pricing today, but institutions should weigh vendor financial durability before committing research or teaching workloads.
- The political risk is now explicit and state-level, so institutions siting their own data centers or negotiating incentive packages should expect longer permitting timelines in several states.
| Read the full story |
AI chip financing hits a wall
The debt markets that fueled the AI data-center buildout are pulling back, with Semafor reporting that some lenders are now trying to unload data-center loans — including ING seeking to offload $500 million of such debt, a slice of which backs Stargate, the OpenAI–SoftBank–Oracle project. The next stress test is a roughly $40 billion chip-financing deal being marketed by Goldman Sachs and Apollo, the largest of its kind, and some major investors are openly declining the category: Ares Management CEO Michael Arougheti said his firm is passing on GPU financing because "I, for one, personally do not know how to underwrite the decline and depreciation curves on that type of technology."
- The core unresolved question for any institution buying AI infrastructure is the same one Ares flagged: nobody has confidence in GPU depreciation curves, so lease-versus-buy decisions for campus AI compute deserve scenario planning rather than a single assumption.
- If debt financing for chip purchases tightens, the compute-supply picture for 2028 and beyond gets more uncertain — one more reason universities should avoid architectural lock-in when planning research and teaching AI capacity.
- Financing stress historically arrives before construction stops; institutions with time-sensitive AI programs should sequence critical capacity commitments earlier rather than later in their roadmaps.
| Read the full story |