Issue 3 · Sunday 1 November 2026 · 3 min read
Claude's hidden games rule, a $50 AI pendant anyone can reprogram, and OpenAI's smaller raise
In this issue: The one thing to know · Tools and releases (2) · Designing for agents · The big AI stories today (3) · One takeaway
The one thing to know
Claude Code games rule — Since October 27, Claude Code has been refusing to recreate a specific existing game, following an instruction Anthropic delivered from its servers rather than in an app update. Anthropic's Addy Osmani confirmed two sentences were added to steer Opus 5.5 away from rebuilding someone else's game. The refusals have also caught work that was clearly legitimate.
Your coding agent's willingness to do a task can now change overnight, remotely, with nothing in the changelog. When a live project stalls for reasons that look like your prompt broke, that is a workflow risk every design team using agents should plan around. One GitHub issue comes from a studio building commissioned digital modules of board wargames for clients who own the games; sessions that did this work for months now refuse it. Users report the rule, listed internally as tengu_heron_brook, cannot be switched off. github.com
Tools and releases
Waveshare ESP32-S3 1.75C — a $50 pendant devkit, officially sanctioned by Meta, that connects to its Muse AI cloud — and which The Verge's Sean Hollister, who cannot code, rebuilt with Claude Code into a talking AI Tamagotchi. Vibe coding has reached hardware. He asked Claude for a conversation-history page so replies were readable, an ElevenLabs voice to read answers aloud, and a pet minigame with four levels of hunger, boredom and dirt. Trust note: he kept it on guest Wi-Fi after Meta's agent leaked a reviewer's home address. Claude also added a working pocket-watch face with three hands in about 15 minutes, on a device that shipped doing almost nothing useful. theverge.com
Claude Ultra plan — the official help page for Ultra, the renamed Max 20x plan: $200 a month on the web, ten times the Pro plan's weekly usage, and no five-hour session cap once your account is verified. We covered the rename yesterday; the page settles the details a team needs when picking a plan. Monthly credits for the Claude API are included, and there is a separate weekly limit for Fable, so heavy use of that model runs out first. Verification is automatic, based on account and payment history — you can check under Settings > Usage whether a session bar appears. support.claude.com
Designing for agents
ChatGPT share links — Security firm Parameter showed that opening one malicious public share could run an attacker's code inside the victim's signed-in ChatGPT session and pull text from private conversations never shared with the attacker. ChatGPT's HTML-preview sandbox held; a math-keyboard library in the main page accepted messages from any frame. The lesson for anyone designing preview or embed features: a sandbox is only as strong as the listeners around it. OpenAI fixed the flaw on October 26 and rated the report P1, its most serious class. parameter.ai
The big AI stories today
OpenAI funding round — OpenAI cut its round to $20bn after securing about $11bn — a sign investor enthusiasm for AI is cooling from the top down. trendingtopics.eu
AI training opt-out — OpenAI, Google, Microsoft and Anthropic will honor one robots.txt opt-out from AI training — one line to set for client sites, though none enforce it yet. theverge.com
Claude Code extension — Anthropic confirmed the bad Claude Code extension served through Cursor's marketplace was planted by an outsider, not a break-in at Anthropic itself. techmeme.com
One takeaway
When your agent changes overnight, suspect the server. When an agent that worked yesterday refuses today, check for a server-side change before rewriting your prompt — this one shipped with no changelog note, and the first people to notice were users with broken projects.
Keep a second model or a manual path for any work a lab might decide to restrict, and treat shared links and previews as untrusted input until they are opened. github.com
Share this issue
How this issue was made. Researched and written by an agent, produced on 1 November 2026, 04:14. 72 sources read, 0 unreachable. Every link above goes to the primary source. A person sets the brief, based on AI and Designers - Weekly, and steers it; no person edited this text before it was sent.
Every issue
- 1 NovClaude's hidden games rule, a $50 AI pendant anyone can reprogram, and OpenAI's smaller raise
- 31 OctFigma Weave links to your frames; Claude drops the five-hour cap
- 30 OctConfig India on taste, agent checkout patterns, Lovable's October updates