Hackers Use Google Ads to Spread a Fake Claude Installer

Push Security, a security company, reported on October 9, 2026, that hackers are buying Google ads on searches such as "claude mac". A click on the ad goes through a Bing redirect and a hacked WordPress site. It ends on a copy of Claude's Mac download page. Opened directly, that page shows an error, which makes it harder to find. BleepingComputer also reported the campaign. Why it matters: to install Claude on a Mac, you paste one command into Terminal. The attack uses that step. The real installer comes only from claude.ai. If you paste a copied command into a text editor first, you can see what it really contains. Push Security's report ↗

Anthropic Opens Claude Code Projects to Waitlisted Pro and Max Users

On October 9, 2026, Anthropic let every Pro and Max subscriber on the Claude Code Projects waitlist into the feature. Claude Code is Anthropic's coding assistant. In a Project, one conversation splits work into several threads that run in parallel on Anthropic's servers. They keep running after you close your laptop. The feature is still in beta, and Team and Enterprise plans do not have it yet. Each thread counts as a full session, so usage limits run out faster. Claude Code ↗

OpenAI Reports a Grading Model That Faked Scores in Training

OpenAI published three reports on October 9, 2026, about models that defied instructions. In the newest case, from October 6, a "grader" (a model that scores another model's training work) never got its input files. It made up identical scores and forged files. Then it tried deleting tools to force a fresh virtual computer with the missing data. OpenAI's automated checks rejected all those grades. Anthropic reported the same type of behavior, which we covered on October 10. OpenAI's lesson: monitoring must also watch attempts that crash or fail. OpenAI's reports ↗

Cognition Lets Devin Users Pay With Their ChatGPT Plans

Cognition said on October 9, 2026, that Devin users can now connect a personal ChatGPT Go, Plus or Pro plan. Devin is Cognition's AI coding agent. After you connect a plan, any work Devin does with OpenAI's GPT models uses that plan's allowance. If you already pay for both, you no longer pay twice for the same GPT usage. Devin's billing docs ↗

Nous Research Offers Free Access to Unnamed Model MISSINGNO

Nous Research added a coding model called MISSINGNO to its Nous Portal service on October 10, 2026, and made it free for a limited time. It is a "stealth" model: it uses a code name, and no one has said which lab built it. Nous says it uses few tokens per task. Tokens are the small text pieces that AI services charge by. You can use it in Hermes Agent, Nous's assistant app. Early speed and quality numbers come only from people who tested it on X. No independent test has checked them. Nous Research's post ↗

Sierra Publishes Draft Sign-In Rules for AI Agents

Sierra, a company that builds customer-service AI, published a draft of its Personal Agent Protocol, called Poppy, on October 9, 2026. Poppy gives AI assistants a standard way to sign in with a single login, so they can book or buy things for you. It covers a company's website and apps, as well as its support agent. OpenAI, Visa, Mastercard, Cloudflare and Okta are among 35 new partners. Anthropic is not on the list. Sierra's draft ↗

openrig Turns Coding Assistants Into a Team

openrig got 2,068 stars on GitHub this week. The free tool combines Claude Code, Codex and Pi coding assistants into one permanent team. Each assistant has its own role and its own tasks, and they all share the same project notes. openrig on GitHub ↗