Morning edition — Friday 16 October 2026

OpenAI published the stable Codex CLI 0.163.0 release at 23:30 CEST on October 15, after the afternoon edition. A follow-up sent during a response can steer the agent immediately; interrupting a turn also stops active Code Mode cells and their nested tool calls.

The release notes say saved threads now restore permissions and network approvals while respecting explicit overrides and current managed policy. The network proxy checks permission before DNS lookup, rejects duplicate Host headers and prevents credential aliases from bypassing inspection hooks.

A searchable /config panel applies supported preferences immediately and identifies settings requiring a restart. Independent read-only tools can execute in parallel, and /daemon can restart a managed background server after warning about work it could interrupt.

Anthropic published Claude Code v2.1.301 at 21:22 CEST on October 15. This is a newer software release than yesterday morning's v2.1.300, with a different set of operational fixes.

The release adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS so unattended retry mode can end a request instead of waiting for a usage limit whose reset is too distant. It also fixes quitting removing a worktree created by a WorktreeCreate hook without asking when that worktree may contain unpushed commits.

Grep and Glob now report an error when the underlying search program fails, rather than presenting the failure as no matches. Another fix addresses workflow agents skipping a step while the run still reported success. OpenTelemetry tool-decision events gain classifier and permission-mode details.

GitHub announced general availability of AI Scan for pull requests at 17:04 CEST on October 15. It complements CodeQL by reviewing changed code in languages and frameworks CodeQL does not currently cover, including PHP, Shell/Bash, Terraform configuration and Dockerfiles.

Findings appear alongside CodeQL results with an AI indicator, an explanation and potentially a suggested fix. They are advisory: they do not fail the code-scanning check and cannot block merges through ruleset requirements. Fork pull requests and Dependabot pull requests are excluded.

Public repositories are supported. Private and internal organization repositories require GitHub Code Security or Advanced Security; a Copilot license is not required. GitHub says the feature consumes Code Security AI Credits.

Anthropic announced at 19:20 CEST on October 15 that it had onboarded six employees from independent evaluation nonprofit METR with company hardware, office space and internal access roughly comparable to its own risk-assessment teams.

METR can publish analyses of risk levels, incidents and practices without an Anthropic veto. Anthropic can require narrow, predefined redactions, including customer or privileged information; METR may publicly characterize the redactions and whether it considers them justified.

Anthropic describes three access tiers. Initial engagements may require publication approval, a second tier allows publication with defined redactions, and the third adds employee-like access. The company says other partnerships have more limited access and rights; the METR terms do not automatically apply to every evaluator.

Google announced at 18:00 CEST on October 15 that more than 400 publications worldwide now participate in its News AI pilot. Newly named partners include Axios, CNN, the Financial Times and Forbes, alongside local US publishers.

Google says the licensing program pays participating publications for enhanced content used to develop AI features. Examples include in-depth article overviews on Google News publisher pages and customizable audio briefings, with attribution and direct links.

In an editor's note published at 17:33 CEST on October 15, Current Affairs said its new union contract includes a policy barring AI-generated text and illustrations from the magazine. The fresh event is the public disclosure; the note says the policy was recently adopted and does not establish its exact adoption time.

The restriction is not a ban on every assistive tool. Editor Nathan J. Robinson explicitly allows tools that help writers research, and small artistic operations such as changing a color or adding a texture, while reserving the magazine's published writing and illustrations for human work.

The Hacker News discussion drew support for that distinction. kolinko wrote, "I kind of respect that. Ok using AI for researching and whatnot, and for their core thing which is writing and illustrations using human work." georgemcbay added, "It is not currently great (or even remotely passable) at writing interesting magazine articles." These are two readers' opinions, not an evaluation of every model or a representative survey.