• President Trump Signs Executive Order on AI Innovation and Cybersecurity — The new executive order directs federal agencies to strengthen cyber defenses using AI, expand access to AI-enabled security tools, and establish a voluntary collaboration framework with AI developers — creating the policy backdrop every university CIO needs to track. 🔗 Graph: [AI Governance], [AI Security], [Higher Ed AI], [EDUCAUSE] 📅 Published: 2026-06-25 📰 https://er.educause.edu/articles/2026/6/president-trump-signs-executive-order-on-ai-innovation-and-cybersecurity 📌 Key takeaways: • The EO directs federal agencies to adopt and expand AI-enabled cybersecurity tools across government systems • A voluntary framework for collaboration between federal agencies and AI developers is being established to share threat intelligence • The order signals the administration's intent to tie AI innovation directly to national cybersecurity posture • For higher ed, this creates a compliance and policy benchmark — institutions receiving federal funding or research grants will need to align with the framework's expectations • The EO reinforces the strategic importance of AI Governance as a board-level and CIO-level priority, directly relevant to UCSD's evolving AI compliance posture

• OpenAI and Broadcom unveil LLM-optimized inference chip (Jalapeño) — OpenAI and Broadcom revealed Jalapeño, a custom ASIC designed from the ground up for LLM inference, claiming performance-per-watt matching Nvidia's Blackwell in early testing with deployment targeted by end of 2026. 🔗 Graph: [OpenAI], [LLM Gateway], [AI Security], [San Diego Supercomputer Center] 📅 Published: 2026-06-24 📰 https://openai.com/index/openai-broadcom-jalapeno-inference-chip 📌 Key takeaways: • Jalapeño is a blank-slate ASIC design purpose-built for LLM inference (not training), meaning it's optimized for serving models to users rather than training new ones • Broadcom CEO Hock Tan stated the chip matches Nvidia Blackwell and Google TPU performance per watt in early benchmarks • The chip took just nine months from design to tape-out — an extremely fast ASIC development cycle • OpenAI describes Jalapeño as the "first step in a multi-generation compute platform," signaling long-term commitment to custom silicon • For UCSD's on-prem TritonAI infrastructure at SDSC, this shift toward purpose-built inference silicon could affect hardware procurement strategy as OpenAI's platform evolves

• Before AI, Fix Your Data — Campus Technology argues that most institutions are deploying AI on top of fragmented, outdated, or ungoverned data — and getting confidently wrong answers as a result — making data readiness the overlooked prerequisite for responsible AI adoption. 🔗 Graph: [Data Analytics], [AI Adoption], [Higher Ed AI], [Data Analytics Governance] 📅 Published: 2026-06-25 📰 https://campustechnology.com/articles/2026/06/25/before-ai-fix-your-data.aspx 📌 Key takeaways: • Generative AI's output quality depends entirely on the quality of the underlying institutional data — sophisticated AI on bad data produces sophisticated-sounding wrong answers • Critical institutional knowledge typically lives across five or more systems (SIS, LMS, CRM, financial aid, departmental apps) with little to no governance • Several institutions that deployed AI assistants before fixing data governance saw tools confidently directing students to outdated policies and resources • Data readiness requires cross-campus collaboration — academic affairs, student services, enrollment, research, finance, and IT all have a stake • Directly supports Brett's Data Analytics Governance priority: the article reinforces that treating information as a strategic asset (not an operational byproduct) is the foundation for AI that can be trusted

• BerriAI LiteLLM Security Advisory: 5 New Vulnerabilities Disclosed — Five new CVEs hit LiteLLM over the weekend of June 21-22, including an authentication bypass (CVSS 7.3) that could allow unauthorized access to proxied AI services — making seven CVEs disclosed for LiteLLM in June 2026 alone. 🔗 Graph: [LiteLLM Enterprise], [AI Security], [LLM Gateway], [AI Compliance & Governance] 📅 Published: 2026-06-22 📰 https://threat-modeling.com/berriai-litellm-new-cves-june-2026-authentication-api-key/ 📌 Key takeaways: • CVE-2026-12773 (CVSS 7.3 HIGH) is an authentication bypass in the UserAPIKeyAuth function affecting LiteLLM up to v1.59.8 — could let attackers consume proxied AI services without authorization • Additional vulnerabilities cover API key exposure and authorization weaknesses across versions up to v1.82.2 • All five CVEs plus the CISA KEV CVE-2026-42271 (command injection) are fixed in LiteLLM v1.63.2+ and v1.82.3+ • Recommendation: upgrade immediately, rotate all AI provider API keys (OpenAI, Anthropic, Google, Azure) that passed through vulnerable instances, and audit access logs for suspicious usage • Directly relevant to Brett: LiteLLM Enterprise is the core LLM gateway at tritonai-api.ucsd.edu — this advisory reinforces the need for strict network segmentation, prompt patching, and API key hygiene as part of TritonAI's security posture

• Databricks Data + AI Summit 2026: Agents, Context, and Governance Take Center Stage — At DAIS 2026 (~30,000 attendees), Databricks launched Genie One (GA), Genie Ontology, Agent Bricks, and Unity AI Gateway — all reinforcing CEO Ali Ghodsi's thesis that "AI has a context problem, not an intelligence problem." 🔗 Graph: [Databricks], [Agentic AI], [Data Analytics], [AI Governance] 📅 Published: 2026-06-19 📰 https://atlan.com/know/ai-agent/databricks/databricks-data-ai-summit-2026-announcements/ 📌 Key takeaways: • Genie One reached GA as a self-service agentic analytics tool, and Genie Ontology adds a live context layer that automatically extracts business knowledge from connected systems • Agent Bricks provides composable agent building blocks; Omnigent is Databricks' vision for multi-agent orchestration • Unity AI Gateway and Unity Catalog Metrics reframe data governance as the grounding layer for agentic AI — a concept directly aligned with TritonAI's architecture • The Summit's cross-cutting themes: agentic AI moving from lab to enterprise, AI cost management, and the need for a "context layer" to ground AI agents in trusted institutional knowledge • For Brett: Databricks is a vendor in his network, and the Unity AI Gateway concept parallels the LiteLLM gateway strategy already deployed at UCSD — worth watching as Databricks extends into AI gateway territory

• How Universities Can Manage Vendor Risk After the Canvas Breach — The 275-million-record Canvas breach exposed systemic gaps in higher ed's third-party risk management, with experts calling for vendor vetting frameworks, contractual safeguards, and continuous asset monitoring as AI agents expand the attack surface. 🔗 Graph: [AI Security], [Higher Ed AI], [Canvas LMS], [UC San Diego] 📅 Published: 2026-06-25 📰 https://edtechmagazine.com/higher/article/2026/06/how-universities-can-manage-vendor-risk-after-canvas-breach-perfcon 📌 Key takeaways: • The Canvas breach impacted 275 million records across ~9,000 institutions — the largest educational data breach on record — and was a wake-up call for third-party vendor risk in higher ed • Experts recommend vendor vetting frameworks that go beyond procurement checklists: breach notification timelines, SLAs tied to academic impact, security audit rights, and explicit forensic cooperation requirements • UW-Madison's response — real-time alerts warning faculty not to click links or reset passwords in Canvas — is held up as a model for transparent crisis communication • AI agents create a new attack surface: "Think of AI agents as a super-application that accesses resources — you need to know who has access to which applications and what their access is" • Relevant to UCSD: the article's emphasis on API key management, asset registries, and vendor SLAs maps directly to Brett's AI Security and Enterprise Monitoring priorities

💡 Signal: Three converging themes this week — AI governance is hardening (Trump EO, LiteLLM CVEs, vendor risk post-Canvas), data readiness is being recognized as the actual bottleneck for AI in higher ed (Campus Technology's strongest statement yet), and the infrastructure layer is maturing fast (OpenAI's custom inference silicon, Databricks' agentic stack). For UCSD, the LiteLLM CVEs are the most actionable item — the same gateway technology running TritonAI's model routing just had seven vulnerabilities disclosed this month, making the case for prompt patching and network segmentation.