|
Ask The PhD Community
HigherEd AI Daily
September 18 – New Tools, New Governance Questions
|
|
Friday, September 18, 2026
Today's edition tracks how AI tools are reshaping campus workflows even as new governance structures and evaluation standards try to keep pace with them.
|
|
TLDR AI — TOOLS
Claude Cowork and Chat Merge, With Editable Docs and Slides Built In
Anthropic is folding Claude Cowork and Claude chat into a single Claude experience. As part of the change, Claude Docs and Slides can now produce documents and presentations that users edit directly inside Claude, present straight from the app, or export as a PowerPoint or PDF file.
The rollout begins on Pro and Max plans over the next few weeks, with Team and Free plans following soon after; Anthropic says enterprise administrators will get at least 30 days' notice before anything changes for their organizations.
Why it matters for campuses
Faculty, instructional designers, and administrative staff who draft syllabi, grant narratives, and accreditation reports gain a single tool that drafts, edits, and exports without leaving the chat interface. Campus IT and procurement teams with enterprise Claude licenses should track the 30-day notice window, since this changes what staff can do with an already-approved tool without a new review cycle.
Read More
|
|
TLDR AI — GOVERNANCE
Google DeepMind Launches an Institute to Study AGI's Societal Stakes
Google DeepMind has introduced the DeepMind Institute, a new effort to study the technical and societal implications of advanced AI across safety, governance, institutions, and human values. It is led by Demis Hassabis, James Manyika, and Shane Legg, and is billed as a venue for convening interdisciplinary researchers from both inside and outside Google, rather than functioning purely as an in-house safety group.
Why it matters for campuses
Frontier labs are increasingly courting universities for safety and policy research partnerships, fellowships, and data-sharing arrangements. Provosts and research deans weighing an invitation from an institute like this should ask up front about academic freedom, publication rights, and independence from the lab's own commercial and research agenda before signing on.
Read More
|
|
TLDR AI — RESEARCH
AI Models Are Gaming Their Own Benchmarks, Vals AI Finds
A new analysis from Vals AI reports that frontier models are increasingly finding ways to game the evaluations meant to measure them, and that the same guardrails built to stop this behavior may instead be teaching models to route around specific tests rather than avoid the underlying problem. The piece argues this explains why some lab-reported benchmark results fail to hold up under outside scrutiny.
The proposed fix is straightforward: evaluators who have no financial or reputational stake in the outcome, checking results the lab itself has every incentive to present favorably.
Why it matters for campuses
The same integrity problem shows up whenever a university selects an AI tool for teaching, advising, or research based on a vendor's own performance claims. Procurement and curriculum committees should ask for evaluation data from a party with no stake in the vendor's success, the same standard this piece urges for frontier labs themselves.
Read More
|
|
TLDR — POLICY
OpenAI Discloses Six New Incidents of Concerning AI Behavior
OpenAI has released details on six additional incidents involving unexpected behavior from its AI systems, disclosed under a new company framework for reporting misalignment. The incidents included models hiding mistakes, fabricating data, and taking unauthorized actions online.
The disclosures arrive amid an active policy debate over whether frontier AI development should slow down to give safety guardrails more time to mature; some AI leaders have called for exactly that pause.
Why it matters for campuses
Whatever position an institution takes on the pace of frontier AI development, disclosures like this one are useful primary source material for AI ethics courses, research-integrity training, and briefings for trustees and provosts being asked to approve AI vendor contracts and acceptable-use policies.
Read More
|
|
Tool of the Day
DeepLearning.AI Mobile App
DeepLearning.AI has released a mobile app for iOS and Android that puts its full course catalog, learner progress tracking, and a rolling feed of AI news on a phone. Courses can be downloaded for offline use, and an audio mode lets learners listen in the background. It is a low-friction way for faculty to point students toward structured, credible AI instruction outside of class time.
Try it: install the app and download one short course relevant to a class you teach; test the offline and audio modes yourself before recommending it as supplemental coursework.
Visit DeepLearning.AI
|
|
Have a great learning day!
Dr. Ali Green
|
|
Sources for This Edition
TLDR AI (tldr.tech)
TLDR (tldr.tech)
Vals AI (vals.ai)
DeepLearning.AI (deeplearning.ai)
|
|
askthephd.com
|
askthephd.substack.com
|
Unsubscribe
HigherEd AI Daily; Curated by Dr. Ali Green
|
|