Ask The PhD Community

HigherEd AI Daily

April 23 – When AI Agents Enter Shared Workspaces, Governance Has to Follow

Thursday, April 23, 2026

The week closes with three developments that together carry one clear message for higher education: AI is becoming embedded in the platforms campuses already use, and governance questions are arriving faster than most institutions expected.

The Rundown AI — Tools

OpenAI Rolls Out Shared AI Agents for Teams Inside ChatGPT

OpenAI introduced Workspace Agents in ChatGPT this week; Codex-powered shared bots designed for multi-step team workflows. Unlike individual GPTs from 2023, these agents retain memory, connect to integrated apps like Slack, and can operate on a schedule without active supervision.

The shift from individual to institutional use is significant. Until now, most campus ChatGPT usage was person by person. Workspace agents open the door to department-level deployments: an agent that drafts advising FAQs from uploaded documents, routes student inquiries, or processes repetitive administrative forms autonomously.

Why it matters for campuses

This is the moment when AI in higher ed stops being a personal productivity tool and starts becoming an institutional workflow layer. Before adoption expands, campuses need clear policies on data access permissions, approval chains for autonomous actions, and student-facing disclosure when AI agents are managing parts of a process.

Read More

The Rundown AI — Governance

Anthropic's Restricted Mythos Model Was Accessed Without Authorization

Anthropic's Mythos model; deemed too powerful for public release and restricted to select partners under Project Glasswing; was reportedly accessed by a private Discord group days after its April 10 launch. The group used naming conventions leaked in a separate data breach and a borrowed contractor login to locate and use the system.

This is not primarily a story about one group accessing one model. It is about what the incident reveals: that even restricted, high-capability AI systems can be reached through supply chain weaknesses, vendor access controls, and information leaked in unrelated breaches. For higher education institutions that rely on vendor relationships to deliver AI tools to students and staff, this is a direct governance signal.

Why it matters for campuses

Vendor access to AI systems is now part of the institutional security surface. Campuses should be asking AI vendors not just about data privacy but about model access controls, incident notification timelines, and what happens when vendor credentials are compromised. These are procurement and IT policy questions; not just legal ones.

Read More

Subscription may be required

TLDR AI — Tools

Google Embeds AI Agents Across Gmail, Drive, Docs, and Sheets

Google announced Workspace Intelligence at Cloud Next this week; integrating Gemini-powered agents across the full Google Workspace suite. The most notable addition for educators is natural language spreadsheet building in Sheets: users describe what they want and the system generates formulas, structure, and analysis. Agents can also act across tools; pulling from emails, files, and calendar data in ways that previously required manual integration.

For the many campuses already on Google Workspace for Education; including institutions like the University of Houston that recently expanded Gemini and NotebookLM access to all students and staff; this upgrade means AI agents are now embedded in tools students and faculty use every day.

Why it matters for campuses

Google's move makes AI agents part of the standard workspace for Google for Education institutions. This accelerates the need for updated faculty guidance, clear policies on AI use in graded work, and communication to students about which parts of their workflow now include AI by default.

Read More

TLDR AI — Policy

Microsoft Moves All GitHub Copilot Subscribers to Token-Based Billing in June

Microsoft announced that all GitHub Copilot subscribers will shift from per-seat to token-based billing starting in June. Business customers will pay $19 per user per month with $30 in pooled AI credits; Enterprise customers will pay $39 per user per month with $70 in pooled credits. The change affects institutions with Copilot access across computer science programs, research computing, or institution-wide developer agreements.

Why it matters for campuses

Token-based billing converts predictable per-seat costs into usage-dependent charges. Institutions with GitHub Copilot agreements should review current usage patterns and assess whether existing terms cover the transition or require renegotiation before June.

Read More

Tool of the Day

Typeless

Typeless is a voice dictation tool designed for professionals who write frequently. Speak your draft naturally; the tool converts it to text; and a paired AI writing layer helps you revise without losing your voice. For educators writing course materials, student feedback, memos, or reports; it replaces the blank page with a spoken draft you can shape.

Try it: Open Typeless and dictate one syllabus section or a student feedback response. Keep the dictated draft untouched; then use an AI assistant to revise it based on your comments. Compare both versions to see where your voice was preserved.

Visit Typeless

It is time to move, refresh, and try something new!

Dr. Ali Green

Founder, Ask The PhD

Sources for This Edition

The Rundown AI (therundown.ai)
TLDR AI (tldrnewsletter.com)
OpenAI (openai.com)
Bloomberg (bloomberg.com)
Testing Catalog (testingcatalog.com)
Where's Your Ed At (wheresyoured.at)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green