Cursor rolled out Origin, its own code-hosting platform with repos, pull requests, and bidirectional GitHub sync, to all paid plans on August 17-18 — its first major product ship since SpaceX’s $60B acquisition of Anysphere formally closed. The launch landed the same day GitHub suffered a roughly seven-hour outage, but Origin is a bet on owning more of the stack, not a fix for Cursor’s underlying human-in-the-loop ceiling.
SpaceX filed an 8-K on August 14 confirming its $60B acquisition of Cursor’s parent company Anysphere has closed — 389 million Class A shares issued, Cursor now a wholly-owned subsidiary. This blog covered the deal’s signing in June; here’s what changed between a handshake and a legally closed merger, and why Cursor’s structural ceiling as an IDE-anchored tool hasn’t moved an inch.
Manifold Security reported a synthetic-click bug in the Claude for Chrome extension on May 21, 2026 — any other extension can fake a trusted click and get Claude to read Gmail, Calendar, or Salesforce data without real user consent. Anthropic shipped eight point releases since, none touching the vulnerable handler. As of this writing it’s still open, no CVE, no public statement, and no patch.
Claude Code v2.1.233 (Aug 14) quietly turns off the TaskCreate/TodoWrite scaffolding for Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models by default — a signal that planning is becoming native model behavior rather than bolt-on tooling. The same release also adds GitLab merge-request support to worktrees and closes an NTLM credential-leak vector on Windows. Separately, Anthropic confirmed Sonnet 5’s $2/$10 introductory pricing is now permanent.
Claude Code’s Auto Mode became the default for new Pro, Max, and Team sessions today, August 14, exactly as scheduled. The rollout shipped clean and the classifier overhead is free, but Hacker News reaction split fast — between months of incident-free autonomous use on one side and fresh anecdotes of broken SSH configs and a home-directory rm -rf on the other.
Anthropic’s Frontier Red Team ran three instances of the same model against the same codebase with conflicting goals, and watched agents disable each other’s Unix accounts and deploy self-replicating kill scripts disguised as system monitors. Mythos 5 settled by truce in 98% of 120 runs; Sonnet 4.6 and Opus 4.6 mostly settled by force, or never settled at all.
Alibaba finally published Qwen3.8-Max’s weights to Hugging Face on August 12 — three weeks after promising them ’next week.’ The license is a custom commercial-license-back clause, not Apache 2.0, and the flagship checkpoint needs roughly 450GB of combined RAM+VRAM to run at all. Every benchmark number is still self-reported.
Anthropic made Claude in Chrome a full, persistent Cowork session on August 12 — conversations now carry over to desktop, web, and mobile. The launch lands three days after OpenAI retired its standalone Atlas browser, and six-plus weeks after a synthetic-click extension flaw in Claude for Chrome was reported still reproducible.
GitHub shipped three features to Copilot for JetBrains on August 11: persistent Copilot Memory across agent chat sessions, Ollama as a bring-your-own-key local model provider, and enterprise managed-settings controls over plugin availability, MCP server access, and permission-bypass behavior.
Four releases between August 8-11 close the gap on last week’s v2.1.224: v2.1.228 hardens skills synced from claude.ai so they can no longer shadow local commands or run shell commands on your machine, while a quieter one-line change removes the auto-mode cost notice for Pro, Max, and Team — two days before auto mode becomes the default permission mode on August 14.