Claude for Teachers launches, memory rebuilt from scratch, Honeycomb spotted in Cursor

Anthropic launched Claude for Teachers on July 14, giving verified US K-12 educators free access to premium Claude capabilities. Once verified, educators get the Learning Commons connector — a direct connection to evidence-based curricula mapped to academic standards in all 50 states — plus a library of teaching skills co-developed with Learning Commons around the tasks teachers said mattered most: lesson planning, differentiation, formative assessments, and student feedback. The product targets the strain on under-resourced schools where planning spills into evenings and class sizes prevent individualized instruction. Claude for Teachers enters a crowded space alongside OpenAI's ChatGPT for Teachers, Microsoft Elevate for Educators, Google's AI Educator Series, and Khan Academy's Khanmigo.

Claude's memory system was completely rearchitected. Instead of a single daily summary that Claude updated at the end of each session, memory now works as a set of individual categorized entries that Claude reads and updates during conversations in real time. Each entry is tagged by category — preferences, projects, people, recurring topics — giving Claude more precise, less lossy recall across sessions. The change is live on web and desktop. Users with memory enabled will notice Claude referencing specific past details rather than broad summaries.

A new Monthly Recap feature at Settings › Reflect shows users the topics they spent time on, their most active day and peak hour, and observations about how they work with Claude. Alongside it, Settings › Time and Focus lets users set optional break reminders and quiet hours. Both features are in beta on Free, Pro, and Max plans on the web and Claude Desktop, and require memory to be enabled.

Healthcare organizations on Claude Enterprise plans and the Claude Platform API can now manage HIPAA readiness themselves — no sales or support ticket required. An eligible admin can review the Business Associate Agreement, download the implementation guide, and enable the HIPAA configuration in a single flow directly in the Claude Console or API dashboard. This applies to both Claude Enterprise and the Claude Platform.

A model listed as 'Claude Honeycomb EAP' (Early Access Preview) briefly appeared inside the Cursor code editor this week before being removed, sparking significant community speculation that Opus 5 — potentially code-named Honeycomb — could arrive before the end of July. Anthropic has not commented. The timing aligns with Fable 5's promotional period ending July 19, leading some to speculate Anthropic is holding a major model launch for immediately after the free-access window closes. The Future of Life Institute's Summer 2026 AI Safety Index, published this week, rated Anthropic highest at C+ — above OpenAI and Google DeepMind at C, Meta at D+, and xAI and DeepSeek described as effectively failing.

Claude Code changed behavior for /verify and /code-review: both skills now only run when you invoke them directly with the slash command. Previously Claude could decide to run them on its own during a session. This change gives developers explicit control over when quality checks occur and reduces unexpected interruptions in long agentic sessions.

A broad stability and safety release shipped this week: tighter permission checks across file and shell operations, safer Bash and PowerShell handling, improved background session cleanup, stronger telemetry, and better remote and plugin reliability. Single-segment dir/** hook conditions now match only the immediate directory (write **/dir/** for any-depth matching). File commands using -m/--magic-file or -f/--files-from now require explicit permission instead of being auto-allowed as read-only. SessionStart hooks now correctly report source 'fork' when a session begins as a fork.

Anthropic extended Fable 5 included access three times — from July 7 to July 12, then to July 19 — each announced hours before the previous deadline. The July 19 deadline is described as final. After 11:59 PM PT tonight, Fable 5 usage on Pro, Max, Team, and eligible Enterprise plans will draw from prepaid usage credits at $10 per million input tokens and $50 per million output tokens. The temporary 50% Claude Code weekly rate limit increase also ends tonight. A brief 30-minute billing glitch on July 17 (~18:30 UTC) incorrectly charged some users for Fable 5 during the included period; Anthropic issued refunds and matching credits.

Claude experienced a service disruption on July 14 with reports spiking around 5:15 PM Eastern Time across claude.ai, the Claude API, and Claude Code. The incident was shorter and smaller in scope than the June 23 outage (which peaked at over 7,000 Downdetector reports). Anthropic's status page logged the incident; no root-cause analysis was published. This is the fourth notable reliability event in June–July 2026.