Ask The PhD Community

HigherEd AI Daily

May 13 – Canvas Breach Tests Vendor Trust in Higher Ed

Wednesday, May 13, 2026

Today's higher ed AI agenda centers on the Canvas breach fallout, Google's deeper embedding of Gemini across Android devices, and a new model class built for real-time classroom-style interaction.

TLDR — GOVERNANCE

Instructure Pays Ransom After Canvas Breach Compromises 275 Million User Records

Instructure has confirmed it paid an undisclosed ransom to the ShinyHunters group following two breaches of its Canvas learning management system. The incidents exposed personal data tied to roughly 275 million users across nearly 9,000 institutions; service disruptions hit campuses that rely on Canvas as their primary academic platform. Instructure says the attackers provided a guarantee that the stolen data was destroyed, although security analysts remain skeptical; ransom payments rarely produce verifiable destruction, and they historically invite repeat targeting.

The breach has drawn unusual scrutiny because Canvas sits at the center of so much of the modern academic enterprise; grades, accommodation records, identity attributes, and instructor communications often live inside the same environment. Several institutions are reportedly reviewing their incident response plans, vendor contracts, and the breadth of personally identifiable information that flows into the LMS by default.

Why it matters for campuses

For higher ed leaders, this is less a one-time vendor incident than a stress test of every institution's third-party risk posture. Provosts, CIOs, and registrars should expect renewed pressure from boards and faculty senates around data minimization in the LMS, contractual breach-notification timelines, and whether sensitive records such as disability accommodations, mental health flags, and advising notes belong inside a tool that is now a confirmed target.

Read More

THE RUNDOWN AI — TOOLS

Google Embeds Gemini Intelligence Across Android and Launches AI-Native Googlebook Laptops

Ahead of next week's I/O event, Google previewed a substantial reorganization of its consumer AI stack. The company introduced Gemini Intelligence as a cross-device platform that performs agentic tasks within apps, reads on-screen context, and links Android phones, tablets, and a forthcoming line of Googlebook laptops built with Dell, HP, Lenovo, Acer, and Asus. The new Googlebooks ship this fall and merge ChromeOS, Android, and Gemini into a single environment that runs Android phone apps natively.

Alongside the hardware push, Google introduced a Gemini-infused mouse pointer billed as Magic Pointer, a Rambler dictation tool that strips filler words, a Create My Widget generator, and Gemini auto-browsing inside Chrome. The thrust of the announcement is that AI is no longer a feature inside a specific app; it now mediates the operating system itself.

Why it matters for campuses

Students arriving with Gemini-native devices will treat AI assistance as the default, not an opt-in. Faculty, IT, and academic integrity offices should anticipate a wave of new questions about what counts as use of generative AI when the operating system itself is generating text, summarizing screens, and acting agentically on a student's behalf. Course-level AI policies that assume a clean line between writing tools and AI tools will need revision.

Read More

THE RUNDOWN AI — RESEARCH

Thinking Machines Lab Debuts Real-Time Interaction Models for Live Human-AI Collaboration

Mira Murati's Thinking Machines Lab released a research preview of interaction models, a new class of system designed for live, multimodal collaboration across voice, video, and text. The model processes input in 200-millisecond chunks, which lets users talk, show, interrupt, and redirect while a second background model handles slower reasoning, search, and tool use. The result is a system that can react to visual changes, count physical repetitions, translate live speech, and speak up at timed moments rather than waiting for a turn.

Murati framed the release as a deliberate counter to the agentic-first direction much of the industry is racing toward; she argued that the way humans work with AI matters as much as how capable the underlying model is. The lab has been quiet since its founding, and this preview is its first significant public differentiation.

Why it matters for campuses

Real-time interaction models open up classroom and laboratory scenarios that prior turn-based chat systems struggled with: live tutoring that responds as a student writes or speaks, accessibility support for deaf and hard-of-hearing learners during synchronous classes, lab observation tools that can comment as procedures unfold, and language practice with a partner that interrupts and corrects in context. Instructional designers should begin asking which pedagogical scenarios benefit from presence and interruption rather than reflective, asynchronous prompting.

Read More

Tool of the Day

Transformers in Practice (DeepLearning.AI)

A new short course from DeepLearning.AI, taught by Sharon Zhou (VP of Engineering and AI at AMD), that opens up the mechanics behind transformer-based models: how text is generated token by token, what attention and positional encoding actually do, why hallucinations occur, and how techniques like RAG, KV caching, and speculative decoding affect inference. Interactive visualizations make concepts that usually live in textbooks tangible enough to teach with.

Try it: Work through the attention and decoding modules before redesigning a class session on AI literacy; even one or two of the visualizations will give you sharper classroom language for explaining to students why a confident model can still be wrong.

Visit DeepLearning.AI

[PLACEHOLDER: Awaiting Dr. Ali Green's 2-sentence closing in her own voice. Please replace this bracketed text with her exact words before sending.]

Dr. Ali Green

Sources for This Edition

TLDR Dev (tldrnewsletter.com)
Inside Higher Ed (insidehighered.com)
The Rundown AI (therundown.ai)
Google (blog.google)
Thinking Machines Lab (thinkingmachines.ai)
DeepLearning.AI (deeplearning.ai)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green