Ask The PhD Community

HigherEd AI Daily

May 30 – Regulation, Risk, and the Uneven Promise of AI Productivity

Saturday, May 30, 2026

As the European Union recalibrates its landmark AI regulations, AI-driven web traffic surges to scale never seen before, and new data reveals that productivity gains from AI tools are concentrating among a small minority of users, higher education leaders have reason to pay close attention to all three developments simultaneously.

The Batch (deeplearning.ai) — Policy

Europe Delays and Weakens Its AI Act Under Industry Pressure

The European Parliament and member states have agreed to amend the EU AI Act, extending key compliance deadlines and softening several provisions after sustained lobbying by businesses and policymakers who argued the law made European companies less competitive. The amendments await formal adoption by the EU Council and Parliament, but the direction of travel is clear: the world's most ambitious AI regulatory framework is being pulled back.

Under the revisions, requirements for AI systems deemed "high-risk" — including those used in law enforcement, critical infrastructure, employment, migration, and personal identification — are delayed from August 2026 to December 2027. Developers would also gain until August 2027 to implement supervised sandbox environments for new model testing. Watermarking and other AI transparency requirements are pushed to around December 2026. The amendments also adjust how personal data may be used in AI training, allowing it for bias detection and mitigation under conditions previously prohibited. Notably, the revisions add one stronger protection: a ban on generating sexually explicit images of children and non-consensual nude images of real people.

Critics have questioned whether loosening the law diminishes its protective function, while supporters argue the changes make compliance achievable without sacrificing innovation. The EU framed the amendments as "safer and simpler rules for both citizens and businesses."

Why it matters for campuses

Institutions operating internationally or hosting EU students and researchers should monitor these changes carefully. Universities using AI in admissions screening, academic performance prediction, HR, or student services may fall within the "high-risk" category under the original Act. The extended deadline to December 2027 gives compliance officers and general counsels more runway, but it also signals that AI governance frameworks remain contested and unstable. Campus AI governance committees would be wise to track the final adopted text rather than assume the earlier timeline is still operative.

Read More

The Batch (deeplearning.ai) — Research

AI Agent Web Traffic Grew 80x Last Year, Raising New Campus Security Questions

According to Human Security's 2026 State of AI Traffic and Cyberthreat Benchmark Report — based on over one quadrillion internet interactions observed in 2025 across more than 200 countries — AI-driven internet traffic nearly tripled year-over-year, while fully agentic traffic (AI agents executing browser-style tasks such as browsing product pages, creating accounts, and completing transactions) grew nearly 80 times over the prior year. Overall automated traffic grew more than 23 percent; human traffic grew by around 3 percent.

Of AI-driven traffic, data crawlers accounted for 68 percent (more than double the prior year's volume), scrapers for 32 percent (a 7x increase), and agents for 1.7 percent in December alone. Agentic activity was concentrated on product and search pages (77 percent of interactions), followed by account pages, authentication flows, and transaction completion. OpenAI was responsible for approximately 69 percent of automated traffic; Anthropic for roughly 11 percent. The researchers flagged a significant share of this activity as malicious: scraping deemed competitive or adversarial rose 47 percent, and agent-created account fraud rose 89 percent year-over-year.

The rise in agentic traffic creates a difficult distinction for security teams: legitimate AI agents now perform many of the same browsing and account-creation behaviors previously associated with malicious bots.

Why it matters for campuses

University libraries, course registration systems, research data repositories, and institutional databases are precisely the types of structured, searchable resources that attract both legitimate AI crawlers and malicious scrapers. Campus IT and cybersecurity offices should audit their current bot-detection policies to ensure they can distinguish between authorized AI research agents, vendor crawlers, and adversarial actors. This is not a future concern; at 80x growth in one year, AI agent traffic is already reshaping the threat landscape that campus security teams manage today.

Read More

The Rundown AI — Access

AI Doubles Developer Output, But the Gains Are Going to a Tiny Minority

Cursor released its Developer Habits Report, drawing on its own platform data to show that AI assistance has more than doubled average developer output: lines of code per developer per week rose from 3,600 to 8,600 over the past 18 months. AI agents are doing more end-to-end work, with tool calls up 30 percent in two months and five times more AI-generated changes reaching commits without manual review. Large pull requests involving more than 1,000 lines of code have also become more common.

But the productivity gains are deeply unequal. The top one percent of developers on the platform produce 46 times more code than the median active user, and the gap is widening every month. Cost per agent request also varies nine-fold across models, meaning that teams choosing more expensive models may be incurring substantially higher costs without proportional gains.

This report focuses on software developers, but its findings are consistent with early research on AI adoption across professions more broadly: a small group of early adopters with strong digital skills, high AI literacy, and frequent, structured usage are capturing the bulk of available productivity gains.

Why it matters for campuses

The productivity concentration described in this report should be a serious concern for instructional designers, faculty developers, and academic leaders rolling out AI tools campus-wide. If the pattern observed in software development holds for faculty research workflows, student writing, and administrative operations, then providing AI access is not sufficient: campuses that do not invest in structured AI literacy training and consistent use scaffolding risk widening existing skills gaps rather than closing them. Equal access to a tool is not the same as equal capacity to benefit from it.

Read More

Tool of the Day

ElevenLabs Dubbing Studio V2

ElevenLabs has released Dubbing Studio V2, an AI-powered dubbing tool that translates and re-voices audio and video content across 90 languages while preserving the speaker's original voice characteristics and timing. The tool is designed for content creators, educators, and organizations that need to make video materials accessible to multilingual audiences without the cost and time of traditional localization workflows.

Try it: Upload a recorded lecture or faculty introduction video and use Dubbing Studio V2 to generate a Spanish or Mandarin version; share the result with international students or prospective students in those language communities and gather feedback on comprehension and quality before investing in broader localization.

Visit ElevenLabs Dubbing Studio

Dr. Ali Green

Sources for This Edition

The Batch by deeplearning.ai (deeplearning.ai)
The Rundown AI (therundown.ai)
Human Security 2026 State of AI Traffic and Cyberthreat Benchmark Report (humansecurity.com)
Cursor Developer Habits Report (cursor.com)
ElevenLabs Dubbing Studio V2 (elevenlabs.io)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green