• Previewing GPT-5.6 Sol: a next-generation model — OpenAI unveiled a three-model suite (Sol, Terra, Luna) with stronger capabilities in coding, science, and cybersecurity. At the US government's request, initial access is limited to a small group of trusted partners whose participation is shared with federal agencies before any broader release. 🔗 Graph: [OpenAI], [Agentic AI], [LLM Gateway], [AI Security] 📅 Published: 2026-06-26 📰 https://openai.com/index/previewing-gpt-5-6-sol 📌 Key takeaways: • GPT-5.6 comes in three tiers: Sol (flagship, strongest in all domains), Terra (mid-tier for high-volume work), and Luna (fast, affordable everyday model) — a tiered approach that lets organizations match capability to cost • On Agent's Last Exam in code mode, Sol is the sole model to exceed 50% task completion (50.9%), while Luna edges out the prior generation's flagship — significant jumps in agentic coding performance • In quantitative biology and genomics testing, Sol and Terra achieve higher accuracy than GPT-5.5 and GPT-5.4 while consuming fewer tokens — meaningful efficiency gains in research domains • All three models are designated "High capability" in cybersecurity by OpenAI's Deployment Safety Hub, pushing past prior performance ceilings in vulnerability research and exploitation • Preview is not a self-service program — limited to organizations with an existing OpenAI account representative, coordinated with the US government under the developing cyber Executive Order framework • For Brett: this tiered model structure (Sol/Terra/Luna) could influence TritonAI's gateway routing strategy — matching workload complexity to model capability to control costs

• US government allows Anthropic limited release of Claude Mythos 5 to critical infrastructure defenders — After a government-led safety review, Commerce Secretary Lutnick determined that appropriate safeguards are in place for Mythos 5 — Anthropic's strongest cybersecurity model — to be redeployed to over 100 US organizations operating and defending critical infrastructure. 🔗 Graph: [Anthropic], [Claude], [AI Security], [AI Governance] 📅 Published: 2026-06-26 📰 https://www.cnn.com/2026/06/26/tech/anthropic-mythos-release 📌 Key takeaways: • Mythos 5 had previously penetrated nearly all classified NSA and US Cyber Command systems within hours during testing, according to a Senate Intelligence Committee hearing — raising urgent questions about AI offensive capability • The redeployment is limited to a set of US organizations that operate and defend critical infrastructure — not a general release — and Fable 5 (a less powerful version) remains blocked by the government • Anthropic stated they've been working with the government since the model was paused on June 12 to demonstrate safety guardrails are effective for defensive use cases • The decision split the administration: some officials wanted to keep the model restricted entirely, while Commerce Secretary Lutnick ultimately approved limited defensive deployment • For UCSD: while Mythos is not directly applicable to campus IT security operations, the precedent of government-controlled AI model access is relevant to the evolving AI governance and compliance landscape Brett navigates

• What SUNY's Systemwide AI Policy Means for Public University IT Leaders — New York's State University system passed a binding AI governance policy in May giving its 64 campuses until end of 2026 to establish guidelines covering bias evaluation, student data privacy, and responsible AI use — creating a template and pressure test for other public university systems. 🔗 Graph: [AI Governance], [Higher Ed AI], [AI Compliance & Governance] 📅 Published: 2026-06-26 📰 https://edtechmagazine.com/higher/article/2026/06/suny-ai-policy-higher-ed-it-governance-perfcon 📌 Key takeaways: • SUNY's policy is binding — not advisory — requiring all 64 campuses to establish or update AI guidelines by end of 2026, covering bias evaluation, student data privacy, and responsible AI use standards • IT leaders at SUNY campuses are now devising processes to evaluate AI vendors, implement governance workflows, protect institutional data, and support responsible AI adoption at scale — all within six months • The policy creates a de facto framework that other public university systems (including UC) will likely benchmark against as state-level AI regulations proliferate • Key tension for IT leaders: how to implement centralized AI governance without slowing down departmental innovation and experimentation • Directly relevant to Brett's AI Governance priority at UCSD — SUNY's binding timeline creates a reference point for what comprehensive systemwide AI policy looks like in practice, including the vendor evaluation and data privacy workflows UC San Diego will need to build

• It's not about Anthropic vs. OpenAI anymore — TechCrunch argues the AI landscape has shifted from a competition between frontier labs to a power struggle between the US government and AI companies over which models get released, as the Trump administration exerts unprecedented control over model deployment decisions. 🔗 Graph: [OpenAI], [Anthropic], [AI Governance] 📅 Published: 2026-06-26 📰 https://techcrunch.com/2026/06/26/its-not-about-anthropic-vs-openai-anymore/ 📌 Key takeaways: • The narrative has fundamentally shifted: the most consequential axis in AI is no longer OpenAI vs. Anthropic, but the US government asserting control over which AI models are released, to whom, and under what conditions • Both GPT-5.6 Sol and Claude Mythos 5 are now subject to government approval for release — a pattern that would have been unthinkable 12 months ago • The administration is using a combination of executive orders, Commerce Department review processes, and voluntary coordination with labs to create what amounts to a de facto model release licensing regime • For enterprises and higher ed: this means model availability will become unpredictable — a model previewed today may not be broadly accessible tomorrow, complicating procurement and integration planning • For Brett's model-agnostic gateway strategy: the trend reinforces the wisdom of maintaining multiple model providers (OpenAI, Anthropic, Google, Azure) through LiteLLM — no single model lineage can be relied upon for uninterrupted access

• OpenAI Launches Full-Scale Effort to Patch Open-Source Bugs as It Takes on Anthropic's Mythos — OpenAI revealed an improved GPT-5.5-Cyber model and launched "Patch the Planet," a Daybreak initiative pairing AI-driven vulnerability discovery with human security researchers to fix open-source software bugs at scale — directly competing with Anthropic's Project Glasswing. 🔗 Graph: [OpenAI], [AI Security], [Anthropic] 📅 Published: 2026-06-22 📰 https://www.wired.com/story/openai-launches-full-scale-effort-to-patch-open-source-bugs-as-it-takes-on-anthropics-mythos/ 📌 Key takeaways: • Patch the Planet pairs AI-driven vulnerability discovery (using GPT-5.5-Cyber) with human security researchers to identify and fix real open-source bugs — shifting from finding vulnerabilities to actually patching them • The initiative is part of OpenAI's broader Daybreak cybersecurity push and directly competes with Anthropic's Project Glasswing, which uses Claude Mythos Preview across 150+ organizations • An improved GPT-5.5-Cyber model provides the vulnerability discovery engine, capable of analyzing large codebases and connecting lower-severity issues into realistic attack chains • This marks a strategic shift: AI labs are no longer just building defensive tools — they're competing to become the infrastructure layer for global software supply chain security • For higher ed IT: open-source software is deeply embedded in campus infrastructure; an AI-driven approach to vulnerability discovery and patching could meaningfully reduce the security debt most institutions carry

• Lessons on AI Leadership — An EDUCAUSE podcast conversation examines how higher education leadership is adapting to AI, emphasizing the need to balance risk and experimentation while preparing institutions and students for rapid technological change — with practical guidance for CIOs navigating the tension between governance and innovation. 🔗 Graph: [Higher Ed AI], [AI Governance], [EDUCAUSE] 📅 Published: 2026-06-25 📰 https://er.educause.edu/podcasts/educause-and-the-integrative-cio/2026/lessons-on-ai-leadership 📌 Key takeaways: • The central tension for higher ed AI leaders: how to balance governance (risk management, compliance, ethics) with experimentation (letting faculty and students explore AI's potential) — getting this balance wrong cuts off either innovation or safety • Leaders who treat AI as just another IT project are missing the point — AI is reshaping the institutional value proposition, not just the technology stack • Successful institutions are creating "safe experimentation zones" — sandboxed environments where faculty can test AI tools without violating data privacy or security policies • The podcast emphasizes that AI readiness is fundamentally about organizational culture change, not technology procurement — the hardest work is getting people to adopt new workflows, not deploying the models • For Brett: this maps directly to the TritonAI Developer API Program and campus enablement strategy — the governance-vs-experimentation tension is the exact challenge the Developer API gateway model is designed to solve

💡 Signal: The defining story this week is the US government's direct control over frontier model releases. OpenAI's GPT-5.6 Sol and Anthropic's Claude Mythos 5 are both being parceled out under government oversight — a structural shift from "market competition" to "government-mediated access" that every enterprise AI strategy needs to account for. For higher ed, SUNY's binding AI governance policy creates a template that the UC system will inevitably benchmark against. And OpenAI's Patch the Planet signals that AI labs are competing to become the infrastructure layer for global cybersecurity — a development with long-term implications for how institutions manage their software supply chain risk.