|
Ask The PhD Community
HigherEd AI Daily
June 6 – AI Governance at an Inflection Point
|
|
Saturday, June 6, 2026
Today's edition takes stock of three converging signals that higher education leaders should watch closely: AI systems beginning to accelerate their own development, a new White House executive order setting guidelines for frontier labs, and the CEOs of the world's leading AI companies urging Congress to act on biosecurity risks that touch directly on research universities and academic science.
|
|
The Rundown AI — GOVERNANCE
When AI Builds Itself: Anthropic Publishes Report on Recursive Self-Improvement
Anthropic published a detailed report this week titled "When AI Builds Itself," examining the early onset of recursive self-improvement (RSI), a process in which AI systems help design and develop their own successors with diminishing human involvement. The report draws on internal data showing that more than 80 percent of Anthropic's merged production code was authored by Claude as of May 2026, with engineers shipping approximately eight times as much code per day in Q2 2026 as they did in 2024.
The report stops short of declaring RSI fully realized, but the authors argue the trajectory is clear: "Each new version of Claude could be built by the version before it, without human involvement," co-author Jack Clark wrote of where current trends lead. Anthropic says it would slow or pause frontier AI development if peer labs agreed to do the same, and plans to convene policy discussions in the coming months to address research protocols, deployment scenarios, and governance structures. OpenAI's concurrent "Democratic Governance of Frontier AI" blueprint independently flagged the first signs of RSI in current systems.
The pace of change is striking: what was framed as a speculative scenario two years ago now has an internal engineering dashboard behind it.
Why it matters for campuses
Universities that have framed AI governance as a medium-term planning exercise are running out of runway. If AI research and development cycles are compressing this rapidly at frontier labs, academic institutions need governance frameworks, faculty AI policies, and research integrity guidelines in place now, not after the next model cycle. Provosts and faculty senates should treat Anthropic's timeline as a planning assumption, not a distant hypothetical.
Read More
|
|
The Batch (deeplearning.ai) — POLICY
White House Issues Executive Order on Frontier AI Models and Cybersecurity
The White House issued a new executive order this week establishing guidelines for companies that build frontier AI models, with a particular focus on cybersecurity implications. The order promotes continued AI development while creating a framework for frontier labs to voluntarily share models with the government for collaborative cybersecurity evaluation. It also mandates increased investment in defensive capabilities to counterbalance advances in automated vulnerability detection.
The order was partly driven by the capabilities demonstrated by Anthropic's Mythos model, which showed substantially improved ability to detect vulnerabilities in code. Andrew Ng, writing in The Batch, described the outcome as "a reasonable compromise between encouraging AI development and protecting security," crediting advisors including David Sacks and Sriram Krishnan for keeping the order from becoming overly burdensome for model developers. Ng also acknowledged that legitimate security risks now give proponents of heavy regulation a more powerful argument than the speculative extinction scenarios that dominated earlier regulatory debates.
The voluntary model-sharing framework and increased emphasis on defensive investment signal that the federal government is attempting to build ongoing relationships with frontier labs rather than impose static compliance requirements.
Why it matters for campuses
Research universities that develop or fine-tune AI models, or that access frontier models through federal grants, should monitor how voluntary model-sharing frameworks evolve into formal requirements. Academic AI research centers and computer science departments may need to revisit their own policies on model development, deployment, and external sharing as federal guidelines take shape. Chief information security officers should also note that automated vulnerability discovery is now a live capability, not a projected one.
Read More
|
|
The Rundown AI — RESEARCH
AI Lab CEOs Sign Open Letter Warning Congress That AI Can Enable Bioweapon Design
The CEOs of OpenAI, Anthropic, Google DeepMind, Scale AI, and Microsoft signed an open letter to Congress this week urging lawmakers to require synthetic-DNA and RNA sellers to screen every buyer and order, verify customer identities, and maintain auditable sales logs. The letter argues that AI has already begun to erode the knowledge barriers that historically made bioweapon development inaccessible to bad actors. "AI systems now outperform PhD-level virologists about highly technical lab procedures in their own domains of expertise," the signers wrote.
Signatories include Sam Altman, Dario Amodei, Mustafa Suleyman, Alexandr Wang, and Demis Hassabis, along with leaders from the DNA-synthesis industry itself. The letter is notable for uniting executives who are direct competitors, and it targets a specific, actionable regulatory gap rather than calling for broad AI oversight. The proposed requirements are aimed at commercial suppliers of biological materials, not at AI labs or research institutions directly.
The letter positions biosecurity as the near-term safety concern most likely to receive bipartisan legislative attention in the current Congress.
Why it matters for campuses
Life sciences programs, biosafety officers, and academic research administrators at institutions with synthetic biology or wet-lab research should track the legislative response to this letter closely. If Congress acts on supply-chain screening requirements, research procurement and vendor compliance procedures may need updating. More broadly, the letter reinforces the case for expanding AI literacy training in biology programs so that students and researchers understand the dual-use risks embedded in tools they are already using.
Read More
|
|
The Rundown AI — TOOLS
ChatGPT's "Dreaming" Update Doubles Factual Recall and Preference-Following
OpenAI introduced a significant memory overhaul for ChatGPT called "dreaming," a background synthesis process that converts past conversations into a continuously updated, category-organized profile of each user. Rather than storing one-off facts, the system builds running summaries across areas including work, travel, learning preferences, and communication style; these summaries update automatically over time and can be reviewed, corrected, or restricted by the user.
OpenAI's internal evaluations show factual recall rising from 41.5 percent to 82.8 percent with dreaming enabled, and preference-following improving from 31.4 percent to 71.3 percent. The feature is rolling out first to Plus and Pro subscribers in the United States, with broader availability to follow. Users retain the ability to review their full memory profile and instruct ChatGPT to avoid certain topics entirely.
Why it matters for campuses
This update has direct implications for how students and faculty experience AI tutoring and writing assistance tools. A system that remembers a student's prior work, learning gaps, and communication preferences provides meaningfully different support than a stateless chatbot. Instructional designers and academic integrity officers should understand how persistent memory changes the nature of AI-assisted work; students who use the same ChatGPT account for coursework across semesters are now working with a tool that accumulates a detailed knowledge of their academic history.
Read More
|
|
Tool of the Day
Perplexity Deep Research
Perplexity Deep Research is a free, browser-based tool that runs multi-step research queries and returns comprehensive, sourced reports within five to ten minutes. It is well-suited for educators who need a rapid landscape review of an emerging topic, a literature scan for course development, or a structured overview of a policy area for grant writing or administrative planning. The free plan includes five deep research queries per day, with no account required to get started.
Try it: Open Perplexity, switch to Deep Research mode, and run the prompt: "What are the leading institutional policy frameworks that universities have adopted for generative AI use in coursework, and what are the key points of disagreement among them?" Walk away for five minutes and return to a sourced comparative analysis you can use directly in a faculty senate discussion or AI task force briefing.
Visit Perplexity Deep Research
|
|
Have a great learning day!
Dr. Ali Green
|
|
Sources for This Edition
The Rundown AI (therundown.ai)
The Batch by DeepLearning.AI (deeplearning.ai)
Anthropic Institute (anthropic.com)
OpenAI (openai.com)
|
|
askthephd.com
|
askthephd.substack.com
|
Unsubscribe
HigherEd AI Daily; Curated by Dr. Ali Green
|
|