- Unexpected AI Agent Activity Reported by Nearly Half of Organizations in Recent Survey — Campus Technology covers Salt Security's State of Agentic AI and API Security 2H 2026 report, which found that 49.8% of surveyed security and technology professionals confirmed or suspected that an AI agent in their environment took an unintended, unexpected, or unauthorized action over the past year. The report also notes agentic AI is now deployed in 86% of organizations, while only a small fraction say they can trace what their agents are actually doing.
📌 Key takeaways:
- Campus IT and security teams deploying AI agents should assume unmonitored agent behavior is already happening — visibility tooling needs to precede agent rollout, not follow it.
- The gap between agent deployment (86% of organizations) and the ability to trace agent actions (a small minority) is the core governance risk this report quantifies.
- For universities, the same pattern applies to both administrative automation and research agents acting against institutional APIs and data stores.
🏛️ UCSD angle: UCSD's campus AI platform routes model traffic through a central LiteLLM enterprise gateway, which gives IT a single point of logging and control over what AI tools and agents can call — directly addressing the agent-visibility gap this survey flags.
- How Higher Ed Institutions Are Managing Third-Party Risk — EdTech Magazine examines how universities are handling risk from the third-party software and SaaS platforms embedded in teaching, research, administration, and student services. The piece points to recent attacks on learning management systems and other platforms as evidence that a vulnerability at a technology provider quickly becomes the institution's problem, and walks through how higher-ed IT leaders are rethinking vendor assessment.
📌 Key takeaways:
- Higher-ed security teams should treat vendor and SaaS risk as institutional risk: an LMS or student-services breach is a campus breach regardless of who hosts it.
- AI features being bolted onto existing third-party platforms expand that attack surface faster than most contract review cycles can keep up with.
- Institutions with structured vendor risk programs — contract clauses, access reviews, and incident-response expectations written into agreements — are absorbing provider incidents better than those relying on trust.
🏛️ UCSD angle: UCSD built an AI contract reviewer as part of its campus AI governance program to scale review of AI-related vendor terms and data-handling clauses — a direct institutional answer to the third-party AI risk problem this article describes.
- OpenAI models are coming to Google Cloud — OpenAI announced that its frontier models and Codex workflows are coming to Gemini's Model Garden, available through Google Cloud Marketplace, with availability rolling out in the coming weeks. The arrangement means enterprises and universities already running on Google Cloud will be able to deploy OpenAI models alongside Google's own and other providers' models from a single marketplace.
📌 Key takeaways:
- Frontier models are increasingly available on whichever cloud an institution already runs on — campus teams should weigh multi-model marketplaces before signing single-vendor AI commitments.
- For research universities with cloud credits or enterprise agreements, this lowers the barrier to comparing OpenAI, Google, and other models on identical workloads and data controls.
- The pattern reinforces that model availability is no longer a lock-in argument; architecture choices should assume the model layer will keep churning.
🏛️ UCSD angle: UCSD's model-agnostic AI gateway (LiteLLM, routing to Claude, GPT, Gemini, Azure OpenAI, and Bedrock) was built on exactly this assumption — new model arrivals like this one slot into the campus platform without re-platforming.
- Employers Say AI Will Make College Degrees More Important — Inside Higher Ed reports on a new Lumina Foundation-Gallup survey of 2,500 U.S. hiring decision-makers finding that 48% believe college degrees will become more important over the next five years amid growing AI use, while only 26% think they will become less important. Employers also expressed more confidence in higher education than the general public did.
📌 Key takeaways:
- This is counter-evidence to the "AI kills the degree" narrative: nearly half of employers see degrees becoming more valuable, not less, as AI spreads through the workplace.
- For institutional strategists, the data supports investing in AI-integrated curricula and credentialing — employer demand appears to be shifting toward degrees that demonstrate judgment AI cannot replicate.
- Messaging and enrollment teams can use employer-side confidence of this kind to counter public skepticism about the value of higher education.
- Could Tech Sovereignty Free Higher Ed From Big Tech? — In an Inside Higher Ed editor's note, Sara Custer argues that greater autonomy over data, infrastructure, and tech stacks could give higher education more say in where AI is going — while acknowledging it is not a solution without risks. The column frames tech sovereignty as a strategic question for institutions that have spent decades consolidating on a handful of large vendors.
📌 Key takeaways:
- Technology leaders should read sovereignty not as building everything in-house, but as preserving real choices over where data and AI workloads run.
- Institutions that control their own gateways, data layers, and integration patterns can negotiate with vendors from a stronger position than those fully consolidated into one stack.
- The risks the column flags are real: self-managed infrastructure carries costs and talent demands many campuses underestimate.
🏛️ UCSD angle: UCSD's approach — an in-house campus AI platform with a self-hosted, model-agnostic gateway rather than a single-vendor AI dependency — is a working example of the middle path this column describes.