• Anthropic Launches Lower-Cost Claude Sonnet 5 — Anthropic released Claude Sonnet 5 as its most autonomous mid-tier model, priced at $2/$10 per million tokens (introductory) — a steep discount to Opus 4.8 — while narrowing the agentic coding gap with its flagship system. 🔗 Graph: Anthropic, Claude, Agentic AI 📅 Published: 2026-07-08 📰 https://campustechnology.com/articles/2026/07/08/anthropic-launches-lower-cost-claude-sonnet-5.aspx 📌 Key takeaways: • Claude Sonnet 5 became the default model for Anthropic's Free and Pro plans on July 1 and is also available through Claude Code, the Claude Platform, and the API as claude-sonnet-5 • Introductory pricing runs $2/M input tokens and $10/M output tokens through August 31, then settles at $3/$15 — still well below Opus 4.8 at $5/$25 • Anthropic says the model can plan multi-step tasks, operate browsers and terminals, and complete agentic work that previously required larger, more expensive models • Early-access partners reported the model completing tasks that earlier Sonnet versions abandoned partway through, including self-checking its own output without being prompted • Ships with cyber safeguards enabled by default; Anthropic published a full system card with detailed safety and capability evaluations • Lower hallucination, sycophancy, and misuse-cooperation rates than Sonnet 4.6, though it still trails Opus 4.8 and Claude Mythos Preview on certain alignment evaluations

• 4 Critical Security Considerations for AI in Higher Education — As generative AI transforms campus operations, CDW analysts outline the four security guardrails every higher-ed CISO and CIO should have in place, starting with data governance and identity and access management. 🔗 Graph: AI Security, AI Governance, Higher Ed AI, UC San Diego 📅 Published: 2026-07-08 📰 https://edtechmagazine.com/higher/article/2026/07/4-critical-security-considerations-ai-higher-education 📌 Key takeaways: • Data governance is the cornerstone: AI systems are only as safe as the data they can access — the principle of least privilege applies to AI queries just as it does to human users • Treat AI like a superuser for IAM purposes: it can query multiple systems and correlate data faster than any human, so it must be governed like an administrator • Institutions must configure AI tools to limit information that would not normally be accessible without AI support — a student query should only return data they are authorized to see • Without proper IAM for AI, institutions risk unauthorized access, data exfiltration, and compliance violations — especially relevant for UC San Diego's 73,000+ user TritonGPT deployment

• Anthropic Launches Claude Cowork on Mobile and Web — Claude Cowork, Anthropic's persistent AI agent platform, expands beyond desktop to iOS, Android, and web with cloud-based processing that keeps tasks running even when your laptop is closed. 🔗 Graph: Anthropic, Claude, Agentic AI, Enterprise AI 📅 Published: 2026-07-07 📰 https://www.theverge.com/ai-artificial-intelligence/961978/anthropic-claude-cowork-mobile-web 📌 Key takeaways: • Claude Cowork is rolling out first to Max subscribers on mobile and web, with broader availability "in the coming weeks" for other plan tiers • Cowork sessions now run in the cloud by default, enabling seamless continuation across devices and background task execution even when no device is online • Scheduled tasks execute autonomously regardless of device status, and Claude can push notifications to your phone when it has something ready for review • The "full experience" — including local file access — remains on the desktop app, but mobile/web unlocks persistent agentic work from anywhere • Doubled Cowork usage limits extended through August, signaling Anthropic is investing heavily in agent adoption

• This Former DeepMind Exec Thinks the AI Arms Race Could End in Disaster — Verity Harding, who led global public policy at DeepMind from 2016–2020, argues that the "AI arms race" narrative is steering the world away from the international cooperation needed to govern the technology safely. 🔗 Graph: AI Governance, AI Security, OpenAI, Anthropic, Google 📅 Published: 2026-07-08 📰 https://www.wired.com/story/verity-harding-ai-arms-race-dangers-anthology/ 📌 Key takeaways: • Harding's new essay anthology, Reframing the AI Arms Race, argues the war-and-weapons language used to describe AI sets the wrong tone for policymaking between nations • She traces the shift from international cooperation (which characterized AI research during her DeepMind tenure) to a rivalries-driven dynamic between labs and superpowers • The Trump administration's nationalist AI rhetoric and export control policies are cited as symptoms of the arms-race framing taking hold at the highest levels • Harding warns that smaller, technology-importing countries are forced to align with one superpower or another, potentially against their own interests • The piece serves as a counterpoint to the competitive frenzy — relevant context for any institution, including UC San Diego, navigating the tension between rapid AI deployment and responsible governance

💡 Signal: This week is dominated by the agentic AI platform race — Anthropic alone shipped two major agent-facing releases (Sonnet 5 pricing and Cowork expansion). The EdTech Magazine security piece is a timely read for anyone running production AI in higher ed: the same week Brett's TritonAI crosses new milestones, the industry is publishing its playbook for doing it safely. Harding's WIRED essay is a useful strategic signal — as university leaders push to deploy AI, the governance question isn't just about compliance, it's about the global framework we're building.