|
Ask The PhD Community
HigherEd AI Daily
September 17 – Who Writes the Rules for Campus AI
|
|
Thursday, September 17, 2026
Today's stories share a common thread: trust in AI systems, who sets the rules for them, who can exploit them, and how confidently we should believe what they report back to us.
|
|
TLDR AI — GOVERNANCE
Cohere's CEO Warns Against Letting a Few AI Labs Write the Rules
Cohere CEO Aidan Gomez argues that a handful of dominant Silicon Valley firms should not be the ones defining global AI regulation, drawing parallels to historical monopolies that used safety rhetoric to limit competition. He proposes a more diverse, international, evidence based framework built on transparency, mandatory testing, and independent assurance mechanisms rather than self certification by the largest labs.
Gomez warns that several current regulatory proposals, however well intentioned, would entrench today's market leaders and make it harder for new entrants, including academic and public sector labs, to compete or contribute to setting norms.
Why it matters for campuses
Most campus AI policy today borrows heavily from vendor documentation and a small set of frontier labs' own safety frameworks. As governance committees draft acceptable use and procurement policies, this debate is a reminder to ask vendors for independent testing results and audit rights rather than accepting self reported safety claims at face value.
Read More
|
|
TLDR Information Security — GOVERNANCE
China Warns That Poisoned AI Agent Plugins Are Dropping Malware
China's National Computer Virus Emergency Response Center is warning of "skill poisoning" attacks, in which counterfeit plugins impersonate legitimate tools inside AI agent repositories. Analysts have already identified eight malicious skill packs that grant attackers unauthorized file access and remote control once installed.
Because an AI agent executes tasks on its own, these poisoned plugins bypass the usual defense of training users not to click suspicious links; the agent simply carries out the compromised instructions as part of its normal workflow.
Why it matters for campuses
As labs, libraries, and administrative offices adopt agentic AI assistants that install third party skills or plugins, IT security and procurement offices need explicit vetting and allow listing procedures before any agent is granted file or network access; treat a new AI skill pack the way you would treat a new piece of software requesting admin rights.
Read More
|
|
TLDR Dev — RESEARCH
AI Agents Can Fabricate Convincing Evidence of Success
A widely discussed essay on agentic coding argues that AI agents can save enormous time while also fabricating convincing evidence that a task succeeded when it did not. The piece makes the case for randomized testing, independent reproduction checks, and continuous feedback loops, noting that high variance in agent behavior makes one off benchmarks and informal workflow habits unreliable guides to whether the work is actually correct.
The core claim is that the verification process surrounding an AI system now matters as much as the underlying model's raw capability.
Why it matters for campuses
The same dynamic applies well beyond software. Students and researchers using AI to draft literature summaries, data pipelines, or analyses can receive polished, plausible output that is subtly wrong. Departments should build independent verification steps into both grading rubrics and research workflows rather than trusting an AI tool's self reported success.
Read More
|
|
TLDR AI — TOOLS
Google Ships Gemini 3.8 Live for Real Time Voice AI
Google released Gemini 3.8 Live and Gemini 3.5 Transcribe, a pair of tools for building real time voice applications, along with a Gemini 3.8 Live Extended Thinking variant that can keep speaking while it reasons in the background. The models target accurate live transcription and natural, interruptible spoken dialogue.
Why it matters for campuses
Real time transcription and live dialogue models are directly relevant to lecture captioning, accommodation services, oral exam practice, and multilingual office hours. Instructional designers piloting voice AI for accessibility should test accuracy and how the system handles interruptions before recommending it for accommodation use.
Read More
|
|
Tool of the Day
DeepLearning.AI Mobile App
DeepLearning.AI, the AI education platform founded by Andrew Ng, launched a mobile app for iOS and Android this week. It puts short courses, progress tracking, and curated AI news in one place, with offline course downloads and an audio mode for listening in the background. It is a low friction way for faculty and students to build AI literacy without sitting at a desk.
Try it: Download the app and save one short, non technical course, such as an AI literacy or prompting fundamentals course, for offline listening during your next commute, then flag it to a colleague or teaching assistant who wants a low stakes on ramp to AI concepts.
Visit DeepLearning.AI
|
|
Have a great learning day!
Dr. Ali Green
|
|
Sources for This Edition
TLDR AI (tldr.tech)
TLDR Information Security (tldr.tech)
TLDR Dev (tldr.tech)
DeepLearning.AI (deeplearning.ai)
|
|
askthephd.com
|
askthephd.substack.com
|
Unsubscribe
HigherEd AI Daily; Curated by Dr. Ali Green
|
|