• TutorMoments: Do AI tutors know when to help and when to hold back? — Ai2 introduces a replay-based evaluation framework that tests whether LLMs can balance the trade-off between stepping in to help a student and holding back to encourage deeper reasoning, using real math tutoring transcripts. 🔗 Graph: Higher Ed AI, Agentic AI, AI Adoption 📅 Published: 2026-08-07 📰 https://allenai.org/blog/tutormoments 📌 Key takeaways: • TutorMoments uses real one-on-one math tutoring transcripts where experienced teachers flag decision points between helping and pushing students to reason independently, then hands the transcript to an LLM to see what it does as the tutor. • When told only to "tutor well," models tend to over-help — giving too much support and rarely pushing students toward deeper thinking, which has direct implications for AI tutoring tools in higher education. • Explicitly spelling out the help-vs-hold-back trade-off in the prompt improves performance but doesn't close the gap to human tutoring, and models differ widely in how reliably they make that pedagogical call. • Ai2 released the de-identified tutoring dataset, replay pipeline code, and model tutor replays for reproducibility — relevant for institutions building AI tutoring systems like UCSD's scheduling assistant and TritonAI education tools. • Watch for: this framework could become a standard benchmark for evaluating AI tutors, pushing vendors toward pedagogically sound design rather than just answer delivery.

• AI-Enabled Ghost Student Fraud: How IT Leaders Are Fighting Back — Generative AI is fueling a surge in synthetic student identities enrolling to siphon financial aid, costing taxpayers hundreds of millions and forcing higher ed IT to adopt layered identity verification and AI-driven pattern detection. 🔗 Graph: AI Security, Higher Ed AI, AI Governance 📅 Published: 2026-08-07 📰 https://edtechmagazine.com/higher/article/2026/08/ai-enabled-ghost-student-fraud-how-it-leaders-are-fighting-back-perfcon 📌 Key takeaways: • Bad actors are using generative AI to create synthetic identities that mimic genuine student behavior long enough to collect financial aid disbursements, with traditional enrollment verification systems unable to keep pace. • The first Higher Education Fraud Summit in July 2026 drew 800+ attendees including the Department of Education, DOJ, and White House Task Force; OIG investigations have already yielded over $35 million in restitution and recoveries. • AI lowers the cost and skill barrier for fraud — what used to take hours of manual ID creation can now be done via text prompt, and bots can mass-submit applications at scale. • For UCSD: as a large federal aid recipient, the campus needs layered identity verification, behavioral analytics, and AI-driven pattern detection integrated into enrollment workflows — a direct concern for Brett's service desk and identity management teams. • Watch for: federal guidance on minimum identity verification standards for financial aid disbursement, and potential new compliance mandates for institutions accepting federal funds.

• Microsoft, Nvidia Move Enterprise AI Toward Active Cyber Defense — Microsoft's Project Perception and Nvidia's Open Secure AI Alliance signal a shift from protecting AI models to using AI agents as active participants in enterprise cyber defense at machine speed. 🔗 Graph: AI Security, Agentic AI, Microsoft, AI Governance 📅 Published: 2026-08-07 📰 https://campustechnology.com/articles/2026/08/07/microsoft-nvidia-move-enterprise-ai-toward-active-cyber-defense.aspx 📌 Key takeaways: • Microsoft's "Project Perception" is an agentic security platform built around a new "Cyber Stack" that continuously perceives, reasons, and acts — coordinating specialized AI agents to identify vulnerabilities and investigate threats while keeping humans in control. • Nvidia's Open Secure AI Alliance (37 partners) targets agent security across multi-vendor clouds by combining open technologies for workload identity, agent controls, vulnerability scanning, and software supply-chain security. • Both approaches treat AI not just as an asset to secure but as an active participant in defense — a fundamental architecture shift from the current alert-and-human-analyst model. • For UCSD: this aligns with Brett's enterprise monitoring modernization initiative and AI-assisted alert filtering — agentic security could eventually replace tier-1 alert triage in ServiceNow workflows. • Watch for: the Open Secure AI Alliance specifications and whether Microsoft opens Project Perception beyond its own security stack to third-party platforms.

• From asking to doing: How the world is putting ChatGPT to work — OpenAI's first country-by-country usage data shows ChatGPT shifting from answering questions to completing tasks, with work-related "doing" usage dominating and global adoption gaps narrowing. 🔗 Graph: OpenAI, AI Adoption, AI Strategy 📅 Published: 2026-08-06 📰 https://openai.com/index/how-the-world-is-putting-chatgpt-to-work 📌 Key takeaways: • At work, people are more than twice as likely to use ChatGPT to complete a task (writing, coding, analysis) than outside work, marking a shift from query-based to task-based AI usage. • Multimedia is the fastest-growing use case globally at 7.8% of messages, with countries like Brazil and Colombia exceeding 10% — signaling expansion beyond text-only AI interactions. • Usage among people over 35 is rising in nearly every country, with France and Czechia seeing 10+ percentage point increases — the "early adopter only" phase is ending. • The data comes from OpenAI Signals, a new hub from the Economic Research Team covering Free, Go, Plus, and Pro accounts — useful baseline data for institutions evaluating AI adoption strategies like UCSD's TritonAI program. • Watch for: whether OpenAI extends Signals data to enterprise/team accounts, which would give institutional leaders like Brett real adoption metrics to benchmark against.

• Will a Data Center Bring Risk or Reward to Fisk? — Fisk University's $400M data center, part of a $900M master plan for the HBCU, has become a flashpoint in the national backlash against data center proliferation and its disproportionate impact on marginalized communities. 🔗 Graph: Higher Ed AI, AI Governance, UC San Diego 📅 Published: 2026-08-07 📰 https://www.insidehighered.com/news/tech-innovation/artificial-intelligence/2026/08/07/will-data-center-bring-risk-or-reward-fisk 📌 Key takeaways: • Fisk University's Quantum Leap plan includes a $400M, 100,000-sq-ft technology and innovation center housing a high-powered data center, aimed at bringing critical tech access to North Nashville and preparing students for a tech-driven workforce. • A petition with nearly 18,000 signatures calls on Fisk to halt the project, citing health impacts of data centers built for generative AI and historic patterns of harmful infrastructure being placed in Black and working-class neighborhoods. • The conflict illustrates the tension between institutional financial stability and community impact — Fisk has faced financial struggles and accreditation challenges, and the data center represents a potential revenue stream. • For UCSD: relevant to the broader conversation about on-prem AI infrastructure decisions — SDSC hosts TritonGPT, and community impact considerations are part of any expansion of AI computing capacity on or near campus. • Watch for: whether community opposition forces Fisk to modify or relocate the data center, which could set a precedent for other universities planning AI infrastructure in residential areas.

• An AI-supervised remote exam went so badly that 58,000 students must retake it — Mexico's largest university deployed AI-powered webcam proctoring for remote entrance exams, resulting in a 5x spike in top scores and a commission finding that widespread cheating rendered results invalid. 🔗 Graph: Higher Ed AI, AI Governance, AI Security 📅 Published: 2026-08-03 📰 https://arstechnica.com/culture/2026/08/an-ai-supervised-remote-exam-went-so-badly-that-58000-students-must-retake-it/ 📌 Key takeaways: • UNAM's first fully remote entrance exam used lockdown browser and AI-powered webcam proctoring for 160,000 applicants — top scores (100+) surged from 3.5% historically to 16.3%, and 110+ scores from 0.9% to 5.5%. • Students exploited the remote format by positioning monitors outside webcam view to access ChatGPT, hiding earphones, or hiring proxy test-takers off-camera — the AI proctoring couldn't detect these physical workarounds. • A commission recommended a "control exam" administered in person, affecting 58,000 people including those admitted since 2021 based on potentially compromised scores. • For UCSD: a cautionary tale as the campus expands AI-mediated services — AI proctoring and AI-assisted identity verification have clear limits, and remote exam integrity remains an unsolved problem even with AI supervision. • Watch for: increased regulatory scrutiny of AI proctoring vendors and a potential shift back to in-person high-stakes testing, which could reverse pandemic-era digital assessment trends.

💡 Signal: This week's feed converges on a single theme: AI is no longer just answering questions — it's acting on the world, and the governance gap is showing. From ghost students exploiting AI to defraud financial aid, to AI proctoring failing to catch basic cheating, to Microsoft and Nvidia building AI agents for cyber defense, the common thread is that AI's capability to act has outpaced the institutional controls designed to contain it. For higher ed IT leaders, the priority shift is clear: the next 12 months will be spent building governance, identity verification, and security architectures that can operate at AI speed.