|
Ask The PhD Community
HigherEd AI Daily
MAY 24 — Grades, Graduation, and the Ground Shifting Under Campus AI
|
|
SUNDAY, MAY 24, 2026
Today’s edition examines how a landmark Harvard grading vote, a wave of graduation-day protests, and new data on AI tool adoption are converging to force every campus leader toward a clearer institutional position on artificial intelligence.
|
|
THE BATCH (DEEPLEARNING.AI) — GOVERNANCE
Harvard Faculty Vote to Cap A Grades at 20 Percent
In a 458-to-201 vote on May 20, Harvard faculty approved a policy limiting A grades in undergraduate courses to approximately 20 percent of the class, with an allowance for professors to assign four additional A’s as a buffer in smaller sections. The measure takes effect in fall 2027 and directly addresses a years-long grade inflation trend in which 66 percent of Harvard undergraduates earned A’s and 84 percent earned an A or A-minus during the 2024-25 academic year. Faculty also passed a companion policy replacing GPA with average percentile rank for determining internal honors and prizes, signaling a broader reorientation toward relative rather than absolute performance metrics.
Student opposition has been vocal: an earlier survey found 85 percent of Harvard undergraduates opposed the cap. The debate has drawn national attention from scholars and technologists alike. Andrew Ng, writing in The Batch, pushed back directly, arguing that capping achievement contradicts the core purpose of education. His position is that institutions should help every student succeed rather than use scarcity to define academic distinction; he advocates for unlimited retries on assignments and designs that prioritize practice over assessment.
The policy’s opponents note a timing problem: just as AI tools make high-quality written and analytical work more accessible to all students, Harvard has chosen to respond by limiting top marks rather than rethinking what top marks measure. Whether the cap will spread to other institutions depends in part on how faculty governance bodies elsewhere weigh the legitimacy of competitive differentiation against the equity implications of restricting academic recognition.
Why it matters for campuses
Provosts and faculty governance committees at institutions across the country will feel pressure to articulate a position on grading standards as AI-assisted work becomes ubiquitous. The Harvard vote reframes the question from “how do we detect AI use?” to “what are our grades actually measuring?” — a shift that requires a principled institutional answer, not a technical one.
Read More |
|
TLDR AI — ACCESS
Gen Z Graduates Are Not Booing AI — They Are Booing a Shrinking Job Market
At commencement ceremonies across the country in May 2026, students booed speakers who praised artificial intelligence. Former Google CEO Eric Schmidt drew pushback at the University of Arizona; music executive Scott Borchetta was jeered at Middle Tennessee State University after telling graduates that AI was rewriting their industry and they should “deal with it.” The impulse is easy to read as technophobia, but the data points elsewhere: unemployment among recent college graduates ages 22 to 27 stood at 5.6 percent in March, compared to 3.1 percent for all college graduates; entry-level roles made up only 38.6 percent of job postings on ZipRecruiter in early 2026, the lowest proportion in at least three years.
A Gallup survey of 1,500 young people ages 14 to 29 conducted in April found that excitement about AI has dropped from 36 percent to 22 percent over the past year, while anger rose from 22 percent to 31 percent, and anxiety held steady at 42 percent. The class of 2026 is the first cohort whose entire undergraduate career unfolded during the generative AI era; they entered as ChatGPT launched and graduated into a market where ServiceNow’s CEO has publicly forecast 30 percent new-graduate unemployment within two years, and Anthropic’s Dario Amodei has estimated AI could eliminate up to half of entry-level white-collar positions.
The boos are not a rejection of technology; they are a signal that institutions have not adequately prepared students for a labor market in structural transition. Writing in The Next Web, analysts argue that what graduates are rejecting is the gap between how AI has been presented to them in higher education and what they are encountering on the other side of the stage.
Why it matters for campuses
Career centers, academic advisors, and curriculum designers are directly implicated. If the class of 2026 leaves with anxiety and insufficient AI fluency, the reputational and enrollment consequences for institutions will follow. The institutions best positioned to address this are those that move now to embed AI competency into degree requirements and career preparation, not as a module but as a thread woven through the curriculum.
Read More |
|
TLDR AI — RESEARCH
State of AI 2026: Half of All Code Is Now AI-Written, and Claude Surpasses ChatGPT in Developer Preference
The State of AI 2026 developer survey, drawing on 7,258 responses collected between April and May, found that AI-generated code now accounts for 56 percent of respondents’ total output, nearly double the 28 percent recorded in 2025. The shift is most pronounced at the upper end of adoption; the proportion of developers generating more than 75 percent of their code via AI tools grew significantly year over year. The survey captures a profession that has moved from experimentation to integration in less than two years.
On model preference, the survey records a meaningful market shift: while ChatGPT remains the most widely used AI tool by sheer volume, Claude has overtaken it in positive developer sentiment and is the model developers are most willing to pay for directly. Use of agentic coding tools such as Claude Code correlates strongly with higher rates of AI-generated output. The top concerns among developers mirror broader societal anxieties: job displacement, AI use in military applications, and the persistence of hallucinations in production systems.
The survey’s findings align with parallel data from Bloomberg and other outlets showing that Cursor, the AI-native code editor, has reached a $3 billion annualized revenue run rate as of May 2026, signaling that developer appetite for AI-assisted workflows is translating into sustained commercial commitment rather than transient curiosity.
Why it matters for campuses
Computer science and information systems departments face an urgent curriculum question: if half of professional code is already AI-generated, what skills define a qualified graduate? The model-preference data also matters for instructional designers and faculty who specify tools in coursework; the market has moved past defaults, and educators advising students on AI tool choice need to understand the professional landscape they are preparing students to enter.
Read More |
|
THE BATCH (DEEPLEARNING.AI) — GOVERNANCE
Google Confirms the First AI-Generated Zero-Day Exploit Used Against a Real Target
Google’s Threat Intelligence Group confirmed in May 2026 the first documented case of a criminal actor using an AI model to discover and weaponize a previously unknown vulnerability, targeting a two-factor authentication bypass in a widely used open-source web administration tool. The vulnerability was not a pattern-matching find; the AI identified a logical flaw rooted in a hardcoded trust assumption, the kind of reasoning-based discovery that standard automated security scanning tools are not designed to detect. Google worked with the affected vendor to issue a patch before the campaign launched, averting a mass exploitation event.
The GTIG report accompanying the disclosure catalogs an emerging ecosystem of AI-assisted attack methods: malware that rewrites its own code to evade antivirus detection, AI systems that reason about code intent to identify bugs invisible to static analysis, traffic obfuscation networks directed by AI to conceal attack origins, and direct exploitation of insecure AI infrastructure. A parallel report from the AI Safety Institute found that frontier AI models can now reliably execute attacks that would require a skilled human hacker three hours to complete. Forensic evidence in the discovered exploit included telltale signs of AI authorship, including oversized explanatory code comments, an invented CVSS score referencing a non-existent CVE database version, and an unusually clean and symmetric code structure.
The disclosure follows a series of high-profile campus security incidents in early 2026, including the Canvas LMS breach that affected more than 9,000 institutions. The convergence of AI-generated attack capability with a sector still recovering from that event should focus institutional attention on the adequacy of current patch management and vulnerability monitoring protocols.
Why it matters for campuses
Higher education institutions operate a wide range of open-source administrative and academic tools, often with limited dedicated security staff. The emergence of AI-powered zero-day discovery compresses the window between a vulnerability’s existence and its exploitation. Campus CISOs and IT leadership should treat this disclosure as a prompt to audit patch management cadences and to evaluate the security posture of any web-based administration platforms currently in use.
Read More |
|
Tool of the Day
Hermes Agent (Nous Research)
Hermes Agent is an open-source AI agent released in February 2026 by Nous Research that runs entirely on your own server or local machine, with no data transmitted to external services and no telemetry. It integrates with institutional email, Slack, Discord, WhatsApp, Signal, and Telegram; connects to whichever underlying AI model you choose; and builds a persistent memory of lessons learned across sessions, automatically generating reusable procedures from successfully completed tasks. Because all data stays on your hardware, Hermes is one of the few AI agents currently capable of operating within institutional data governance and FERPA compliance constraints without a vendor data processing agreement.
Try it: Set up Hermes Agent on a department or research server and connect it to your institutional email account. Ask it to read the last 30 days of student inquiry messages, categorize them by topic, and draft response templates for the three most common inquiry types. Hermes will save those templates as reusable skills it can apply automatically in future sessions, reducing repetitive administrative load without routing student data through a commercial cloud service.
Visit Hermes Agent |
|
Dr. Ali Green
|
|
Sources for This Edition
The Batch by DeepLearning.AI (deeplearning.ai)
TLDR AI (tldrnewsletter.com)
Inside Higher Ed (insidehighered.com)
The Next Web (thenextweb.com)
State of AI 2026 (2026.stateofai.dev)
Google Cloud Threat Intelligence Blog (cloud.google.com)
CNBC (cnbc.com)
Fortune (fortune.com)
|
|
askthephd.com
|
askthephd.substack.com
|
Unsubscribe
HigherEd AI Daily; Curated by Dr. Ali Green
|
|