Ask The PhD Community

HigherEd AI Daily

June 27 – AI Programs Surge as Federal Oversight Expands

Saturday, June 27, 2026

Today's edition covers the rapid expansion of AI degree programs across U.S. campuses, a federal move that will shape how institutions access frontier AI tools, open-weights research capabilities that can level the playing field for under-resourced labs, and career lessons that matter to graduate advisors and their students.

The Batch (deeplearning.ai) — CURRICULUM

AI Degrees Are No Longer Niche: 1,000-Plus Programs Now Across 584 U.S. Colleges

The Center for Inclusive Computing at Northeastern University reports that as of April 2026, U.S. colleges and universities now offer at least 1,000 AI-related programs spanning 584 institutions. That includes 78 dedicated AI majors and 103 minors. Five years ago, in 2021, only five schools offered AI majors. The growth reflects both student demand and institutional recognition that AI literacy is becoming a foundational professional competency across disciplines.

The programs vary considerably in design. Carnegie Mellon, which launched the first U.S. AI bachelor's degree in 2018, requires seven mathematics and statistics courses alongside AI, ethics, and human cognition coursework. Stanford offers an AI concentration built around seven qualifying courses in natural language processing, computer vision, and robotics. Drake University in Iowa takes a markedly different approach with a bachelor's in AI oriented toward humanities and business students, requiring only two math courses and offering flexible clusters in philosophy, psychology, English, and information systems.

Curriculum designers face a structural tension that The Batch editorial team put plainly: AI is moving so quickly that the pace of academic curriculum change is poorly matched to AI's rapid evolution. Innovative faculty are finding ways to move faster, but the gap between what graduates are learning and what the field requires remains a genuine concern.

Why it matters for campuses

Provosts, deans, and faculty governance bodies are under increasing pressure to answer a straightforward question: what does an AI-competent graduate look like, and who on campus gets to define that? The Drake model, which requires minimal math, signals that AI education need not be siloed inside engineering or computer science; it opens space for interdisciplinary programs that serve the broader student population. Institutions that have not yet launched formal AI coursework are now operating against a rapidly normalizing baseline.

Read More

TLDR AI — GOVERNANCE

White House Asks OpenAI to Delay Next-Generation Frontier Model Over Safety Concerns

The White House has issued an official administrative request asking OpenAI to delay the public deployment of its next-generation frontier model. The stated grounds are national security and structural safety concerns. Government officials are pushing for an extended red-teaming window to audit the system's advanced cyber-capability execution limits and automated social manipulation vulnerabilities before the model becomes publicly available.

The request is notable on several levels. It represents a more assertive posture from the executive branch toward frontier AI releases than has been typical, and it signals that safety evaluation processes currently employed by labs may not satisfy federal expectations for systems with autonomous agentic capabilities. It also follows a period in which access to several frontier models has already been restricted for non-U.S. users, adding to a patchwork of controls that academic communities navigating international research collaborations are beginning to feel directly.

Why it matters for campuses

Academic leaders who have built instructional workflows, research pipelines, or campus AI strategies around specific commercial models need to plan for tool availability disruptions driven by policy rather than product decisions. Chief information officers and academic technology officers in particular should monitor how federal oversight of frontier models evolves; the question of which AI systems can be lawfully deployed in federally funded research contexts is likely to become more complex, not less, over the coming academic year.

Read More

The Batch (deeplearning.ai) — RESEARCH

ESMFold2 Makes Molecular Structure Prediction Free and Open to All Researchers

A team at Biohub and EvolutionaryScale has released ESMFold2, an open-weights model that predicts the three-dimensional shapes of proteins, DNA, RNA, and molecules that bind to them. The model treats biological sequences the way large language models treat text: by embedding amino-acid sequences, base-pair sequences, and molecular descriptions and mapping them to structural coordinates. It is available at no cost through the EvolutionaryScale website, on Hugging Face, and via a Biohub API.

Performance benchmarks show ESMFold2 outperforms AlphaFold 3 and competitors when input data does not include multiple sequence alignments (MSAs), which is particularly relevant for novel or synthetic molecules where related biological sequence information is scarce. For researchers working on rapidly evolving viral proteins or synthetic biology applications, this represents a meaningful capability advantage. The model's 6.2-billion-parameter mixed architecture scored a protein shape accuracy of 0.85 lDDT on FoldBench without MSAs, compared to 0.81 for Chai-1.

The open-weights release means that any research team, anywhere in the world, can download and run the model without paying API fees or navigating institutional data agreements with commercial providers.

Why it matters for campuses

Life sciences faculty at institutions without deep computational biology budgets now have access to a state-of-the-art structure prediction tool that rivals what was previously available only through well-funded labs or expensive licensing arrangements. For graduate programs in biochemistry, pharmacology, and molecular biology, ESMFold2 is a practical addition to research methods courses and independent study projects. It also exemplifies a broader pattern in AI research: open-source releases from academic and non-profit labs are steadily narrowing the capability gap with proprietary systems.

Read More

TLDR AI — RESEARCH

What a Fifth-Year PhD Student Learned From the AI-Era Research Scientist Job Search

A fifth-year doctoral student at Brown University published a candid account of their research scientist job search in Silicon Valley, and the lessons cut against several widely held assumptions in graduate advising. Chief among them: only one or two published research papers actually moved the needle with employers. The volume of publications that might satisfy a dissertation committee or a tenure file had little bearing on industry hiring outcomes. What employers evaluated, the author found, was breadth alongside depth; interviewers frequently tested candidates on topics well outside their dissertation specialty.

Timing, the author notes, had an outsized effect on outcomes; the same candidate could have a very different experience depending on hiring cycles, team headcount shifts, and the pace at which specific research domains were attracting investment. Interview formats also varied far more than candidates typically expect, ranging from deep technical probes to product-oriented case discussions, with little standardization across organizations.

Why it matters for campuses

Faculty advisors and graduate program directors who counsel students entering AI-adjacent industry roles may be working from outdated mental models of what those hiring processes actually look like. Career services offices that support doctoral students in STEM and social science programs would benefit from inviting practitioners to speak directly to the format and substance of current research scientist interview cycles. The author's account also reinforces the case for doctoral students building breadth deliberately, not only through coursework but through reading, seminar participation, and engagement with research areas adjacent to their primary focus.

Read More

Tool of the Day

Memoket AI

Memoket is a wearable AI capture device worn as a wristband, pendant, or Apple Watch attachment that records spoken conversations with a single press and automatically generates summaries, action items, and structured reports. It is designed for professionals who move between meetings and contexts throughout the day and cannot afford to lose information that surfaces in real-time conversation. For educators who hold office hours, attend committee meetings, lead advising sessions, and participate in faculty governance, it offers a low-friction way to capture and organize what was said without requiring a phone on the table or manual note-taking.

Try it: Wear Memoket to your next student advising session or department committee meeting, then use the auto-generated summary to draft your follow-up email or meeting minutes within ten minutes of leaving the room.

Visit Memoket AI

Have a great learning day!

Dr. Ali Green

Sources for This Edition

The Batch by DeepLearning.AI (deeplearning.ai)
TLDR AI (tldrnewsletter.com)
TechCrunch (techcrunch.com)
Center for Inclusive Computing, Northeastern University (northeastern.edu)
EvolutionaryScale (evolutionaryscale.ai)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green