Skip to content

Weekly digest

26 October – 1 November 2026

6 posts

From November 29, OpenAI may use content from Daybreak Blue sessions on personal accounts to improve its models even with the model-improvement setting off, OpenAI told users by email. The user who posted the email says that, as far as they know, Daybreak Blue can’t be switched on or off in the Codex CLI on personal accounts.

OpenCode, the open source coding agent, posted on X that a Claude subscription is now usable inside it, and labeled the news “legit”.

OpenCode says Claude subscriptions work in its agent

The open source coding agent OpenCode posted on X that a Claude subscription is now usable inside it, and labeled the news “legit”. The claim comes from OpenCode’s own account. According to CellCog, Anthropic’s SDK docs bar claude.ai login in third-party tools unless approved.

Hacker NewsDiscuss

Hidden Claude Mythos 5.5 line found on Anthropic’s site

A hidden mention of Claude Mythos 5.5 sits on Anthropic’s official Mythos page, according to a post by the X account kimmonismus, which quotes it as “Our most capable model class for cybersecurity and biology research.” The evidence is one page string spotted by one account. The poster labels the idea of a release next week, with Fable 5.5 alongside, as speculation.

The public page lists Claude Mythos 5 as the latest update to Mythos Preview, with gains in cybersecurity, biology and healthcare benchmarks. It’s available only to a small group of vetted partners.

Testing Catalog ThreadsDiscuss

KITP group posts statement on OpenAI’s Yang-Mills proof

A working group at KITP, UCSB, has posted a statement titled “Proof by OpenAI of the Existence of Quantum Yang-Mills Theory.” The proof itself is still a company claim under expert scrutiny, and the statement is dated October 2026. OpenAI says its internal models produced an existence proof for quantum chromodynamics that includes pure Yang-Mills theory, part of a Clay Mathematics Institute Millennium Prize problem, and announced it on October 26, 2026.

Hacker News CodexDiscuss

OpenAI sets a January 1, 2027 shutdown date for Prism

Prism, OpenAI’s AI web app for scientists, will shut down on January 1, 2027, according to OpenAI’s Prism pages. Per Dealroom, OpenAI confirmed it’s shuttering the app, which launched in January as a web app to help scientists work with AI.

Hacker News CodexDiscuss

OpenAI opt-out skips personal Daybreak Blue sessions

From November 29, OpenAI may use content from Daybreak Blue sessions on personal accounts to improve its models, even with the model-improvement setting off, the company told users by email. If you rely on that setting to keep your work out of training, it won’t cover those sessions after that date.

The user who posted the email says that, as far as they know, Daybreak Blue can’t be switched on or off in the Codex CLI on personal accounts.

Hacker News CodexDiscuss

GitHub Copilot gets five new features at Universe’26

At Universe’26, GitHub announced a batch of Copilot features, led by Copilot Fusion, which coordinates multiple models and workflows when one model alone isn’t enough. If agents already write code in your repos, the Security extension targets that output: it reviews what you and your agents just wrote alongside code scanning alerts in one Security canvas.

• Copilot Automations runs a job you define once on demand, on a schedule, or in response to activity in GitHub or a connected service. • Copilot Projects shows where the work stands and lets you direct what Copilot takes on next. • Copilot dreaming suggests improvements to your personal memories, repository instructions, agent skills and custom instructions.

Testing Catalog ThreadsDiscuss