AlphaSignal — "Anthropic Claude Ultra kills 5-hour cap for verified users" (Nov 2 2026)
Why this is in the vault
Standard Monday multi-item Signals curation (Top News x1 / Top Repo x2 / Signals x6, not the Sunday Deep Dive variant). Kept for the Anthropic plan-rename-plus-cap-removal story, which lands directly on the plan this COO instance runs under, plus a secondary Claude Code supply-chain-attack item worth a standing watch note.
Mapping against Ray Data Co
The concrete connection: per reference_claude_plan_max20x, the founder's own plan is Max 20x — which this issue reports Anthropic has renamed to Ultra ($200/month, same price), with the companion change that the 5-hour usage cap is now removed for verified Ultra accounts (weekly limits unchanged, newer/unverified accounts keep a temporary cap until verified). This is the same thread as [[2026-09-01-alphasignal-claude-code-limits-grok-shopping]], which covered Anthropic's permanent 25%-felt-17%-cut weekly-limit change effective 2026-09-14 — that note flagged session-budget planning for heavy fan-out work (deep-research nightly runs, family-research-round, /check-board) as a direct operational constraint. The 5-hour cap specifically is the one that bites hardest on long single-session work: multi-hour brigade runs, overnight agent fleets, and any session that runs past the old 5-hour wall used to get cut off mid-flow regardless of weekly headroom remaining. Removing it for verified accounts is a genuine capacity unlock for exactly that failure mode — worth confirming verification status is current, since the email notes newer/unverified accounts keep a temporary limit. It does not change the weekly ceiling, so the 09-01 pre-mortem on session-hungry recurring skills still holds; this just removes one specific intra-week wall. Read against [[2026-09-11-every-run-out-of-ai-usage-limits]] — Jack Cheng's argument that hitting usage limits is a useful forcing function against "snacking" — the cap removal cuts the other way: it's a removed constraint, not an added one, so the discipline question (don't let more headroom turn into reflexive over-delegation) is still the founder's call, not something this note resolves.
Secondary, separate item: Cursor's warning that 5,000 users had a fake "Claude Code" VS Code extension planting SSH backdoors on their machines is a live instance of the supply-chain-is-the-attack-surface risk that feedback_mcp_install_security_review_default already codifies as a standing posture (security-review third-party installs before install, even unasked). No new SOP needed — this is confirming evidence the existing posture is pointed at a real, current threat class, not a hypothetical one.
Curation section
Zero deep-fetches — all "READ MORE" links route through AlphaSignal's own tracking redirect rather than a resolvable third-party domain, and the newsletter body carries full detail for each lead item itself.
- Top News — Anthropic renames Claude plans and drops the 5-hour limit for verified Ultra accounts. Max 5x → "Max" ($100/mo), Max 20x → "Ultra" ($200/mo), same prices, no change for existing subscribers otherwise. The 5-hour session cap is removed for verified Ultra accounts (rolling out now); weekly limits unchanged; newer/unverified accounts keep a temporary limit until verified. Presented by Tiger Data.
- Top Repo — a developer rebuilt Apple-adjacent startup Subvocal's silent-speech-to-text feature for AirPods Pro in 9 minutes with one prompt. Subvox (open-source, GitHub) reads AirPods motion-sensor data during silent mouthing and pattern-matches against user-trained word templates — no AI model, no vision model. Framed as evidence that a funded startup's months-long build is now a single-prompt afternoon project. Presented by Pave.
- Top Repo — Vercel Labs released
agent-desktop, an open-source tool that lets AI agents control a whole desktop via the OS accessibility tree (stable element IDs instead of screenshot-and-guess), enabling multiple parallel agent cursors on one machine; works with Claude Code, Codex, any shell-capable agent. - Signals (6 items): Cursor warns 5,000 users of malware (SSH backdoors) hidden in a fake "Claude Code" VS Code extension; Decodo's Web Data API pitched for AI-agent web scraping (Presented by Decodo); Scenario opens agent skills for branded image/video/3D/audio generation; a new method trains foundation models in one unified pass, beating staged training; a red heart emoji reportedly makes Claude and GPT pick recursion 100% of the time in some test; 20 free open-source "motion studio" skills drop for Claude on GitHub.
⚠️ Sponsorship
Two standalone "Presented by" placements (Tiger Data, Pave) plus one native in-Signals-list ad (Decodo). All three are confirmed recurring AlphaSignal rotating-pool members per the process-newsletter README (Tiger Data since 2026-09-04, Pave since 2026-10-11, Decodo since 2026-10-30) — no new pool entrant this issue. The masthead "In Partnership with ·" slot is present but fully unresolved in the extracted plaintext (no name, not even generic alt text), consistent with the majority-unresolved pattern logged across September–November issues. None of the Top News/Top Repo items or the unsponsored Signals items carry a disclosed commercial relationship; the Anthropic plan-rename story and the Cursor malware story both read as pure editorial coverage with no sponsor overlap.
Related
- [[2026-09-01-alphasignal-claude-code-limits-grok-shopping]] — the prior Anthropic usage-limit change (permanent weekly-limit shift, Sept 14) this issue's 5-hour-cap removal continues as the same operational thread
- [[2026-09-11-every-run-out-of-ai-usage-limits]] — the counter-argument that usage limits are a useful forcing function, relevant friction against treating this cap removal as pure upside
- [[feedback_mcp_install_security_review_default]] — the standing security-review posture the Cursor fake-Claude-Code-extension item confirms as live, not hypothetical
- [[reference_claude_plan_max20x]] — the founder's own plan, directly renamed by this issue's headline story