• Higher Ed IT Leaders Must Balance AI Innovation With Secure Core Systems — ListEdTech's 2026 report identifies "the stability paradox": universities must modernize for AI while keeping mission-critical systems like ERP, IAM, and registration running reliably. 🔗 Graph: AI Governance, AI Adoption, Enterprise Monitoring, UC San Diego 📅 Published: 2026-07-17 📰 https://edtechmagazine.com/higher/article/2026/07/higher-ed-it-leaders-must-balance-ai-innovation-secure-core-systems 📌 Key takeaways: • A 2026 ListEdTech report found that the top IT investment priorities at 55 universities are data/storage, IAM, ERP modernization, and network infrastructure — all ranked above AI • CEO Justin Ménard frames the tension: stakeholders want AI capabilities, but "universities still need to keep the lights on" with core systems that cannot fail • The "stability paradox" directly mirrors Brett's portfolio challenge at UCSD — advancing TritonAI while maintaining ServiceNow, identity, and collaboration platforms • Institutions that invested in hybrid cloud flexibility before 2020 pivoted faster to AI; the same modernization principle applies now for agentic AI readiness

• A scorecard for the AI age — OpenAI CFO Sarah Friar introduces a practical framework for measuring AI ROI through useful work, cost per successful task, dependability, and return on compute. 🔗 Graph: OpenAI, AI Governance, AI Strategy, Budget & Recharge 📅 Published: 2026-07-17 📰 https://openai.com/index/a-scorecard-for-the-ai-age 📌 Key takeaways: • OpenAI's CFO published a four-metric scorecard: useful work output, cost per successful task, dependability rate, and return on compute invested • The framework pushes enterprises past vanity metrics (queries, sessions) toward outcome-based measurement of AI value • Directly relevant to UCSD's recharge model design — measuring TritonAI ROI by task completion rather than raw usage will be critical for the consumption-based pricing Brett is building with Finance • The scorecard approach could serve as a template for how ITS reports AI value to cabinet and UC leadership

• OpenAI Staffers Are Funding a Rival Super PAC to Take on Their Boss — Seven current OpenAI employees donated over $215,000 to Guardrails Alliance, a super PAC advocating for stricter AI regulation, countering a $100M+ pro-industry PAC backed by OpenAI president Greg Brockman. 🔗 Graph: OpenAI, AI Governance, AI Compliance & Governance 📅 Published: 2026-07-15 📰 https://www.wired.com/story/openai-employees-donations-guardrails-alliance-leading-the-future/ 📌 Key takeaways: • OpenAI research engineer Juan Felipe Cerón Uribe donated $200,000 to Guardrails Alliance, citing concern that safety research won't translate to real guardrails without regulation • Guardrails Alliance launched with $5M initial funding and aims to raise $15M this cycle; it's backed by tech workers, labor unions, and pro-regulation groups • The rival PAC, Leading the Future, has $100M+ from tech leaders including Brockman and his wife ($50M commitment) • This internal revolt at OpenAI signals deepening tension between AI safety practitioners and business leadership — a dynamic that shapes the governance landscape Brett navigates at UC system level

• HG-RAG: Hierarchy-Guided Retrieval-Augmented Generation for Structured Knowledge Graphs — A new framework that performs graph traversal over hierarchical knowledge graphs to deliver structured context to LLMs, outperforming flat-document RAG on hierarchical, relational, and multi-hop reasoning tasks. 🔗 Graph: TritonGPT, Onyx, Model Context Protocol, Data Analytics 📅 Published: 2026-07-18 📰 https://arxiv.org/abs/2607.14095 📌 Key takeaways: • HG-RAG retrieves context by resolving a named entity anchor from a query, then expanding upward through parent nodes, laterally through relational neighbors, and downward through child nodes • Evaluated against dense retrieval baseline across three world scales (18–800 nodes) with four query types: local fact, hierarchical, neighborhood, and multi-hop • Results show consistent improvement on hierarchical and multi-hop reasoning while reducing hallucination and maintaining locality coherence • The architecture is directly applicable to TritonGPT's RAG pipeline — Onyx currently retrieves from flat document stores, and hierarchical graph-based retrieval could improve enterprise data agent accuracy on complex UCSD policy and data warehouse queries

• AI didn't kill the liberal arts: It's making a new case for them — Rollins College president Brooke Barnett describes how faculty adopted an AI literacy framework that prioritizes human-centered learning while integrating AI fluency across the curriculum. 🔗 Graph: Higher Ed AI, AI Adoption, AI Strategy 📅 Published: 2026-07-17 📰 https://universitybusiness.com/ai-didnt-kill-the-liberal-arts-its-making-a-new-case-for-them/ 📌 Key takeaways: • Rollins College developed an AI literacy framework through a year of AAC&U Institute participation, combining foundational knowledge, practical skills, and ethical/social awareness • Faculty explicitly debated: environmental costs of AI, protecting student independent thinking, trust erosion, and whether some tasks should deliberately avoid AI because "the difficulty is the point of the learning" • The core principle: AI fluency is valued, but the educational experience remains human-centered — AI is paired with student judgment, not substituted for it • This mirrors the TritonAI philosophy at UCSD — vertical AI that augments human work rather than replacing the human judgment loop in academic and administrative contexts

• Security incident disclosure — July 2026 — Hugging Face disclosed that an autonomous AI agent system breached its production infrastructure through a malicious dataset, marking one of the first documented "agentic attacker" intrusions in the AI platform ecosystem. 🔗 Graph: AI Security, Agentic AI, AI Governance 📅 Published: 2026-07-16 📰 https://huggingface.co/blog/security-incident-july-2026 📌 Key takeaways: • The intrusion exploited two code-execution paths in Hugging Face's dataset processing pipeline (a remote-code dataset loader and a template-injection in dataset configuration), then escalated to node-level access and lateral movement across internal clusters • The attack was run end-to-end by an autonomous agent framework operating across thousands of short-lived sandboxes with self-migrating C2 on public services — the "agentic attacker" scenario the industry has been forecasting • Hugging Face detected no tampering with public models, datasets, or Spaces, and verified the software supply chain (container images, packages) was clean • For UCSD's TritonAI stack: this validates concerns about dataset-processing pipeline security and the need for stringent admission controls on any external data ingestion into Onyx or the TritonAI platform

💡 Signal: This week's signal cuts across three axes — the governance gap (OpenAI staffers funding rival PACs while Hugging Face gets breached by an autonomous agent), the measurement imperative (OpenAI's ROI scorecard arriving just as higher ed faces the "stability paradox" of funding AI alongside core systems), and the architectural frontier (HG-RAG's hierarchical retrieval pointing toward the next evolution of RAG systems like Onyx). The agentic attacker breach at Hugging Face is a wake-up call for any institution ingesting external datasets into AI pipelines.