Claude News
FR
Live

Claude, the essentials — edition of October 13, 2026

Anthropic's Cyber Mission targets critical infrastructure and open-source code

By ·

Anthropic has launched a long-term cyber-defense effort that starts with operational technology and open-source software. Alongside it come an updated Usage Policy, new research headlines on model behavior and Claude Code releases that harden how hooks and subagents work.

Key points

A cyber-defense mission built on the lessons of Glasswing

On October 8, Anthropic announced the Anthropic Cyber Mission, which it describes as a long-term commitment to securing the systems everyone depends on. It begins with two areas. The first is critical infrastructure, meaning the operational technology behind power grids, water systems and transportation networks, plus government systems. There, the new Critical Infrastructure Defense Program brings frontier models, on-site engineers and threat research to the defenders who protect those systems. The second is open-source software, where Anthropic is launching OSS Scanner, which offers open-source projects regular security scans from its strongest models, for free. A separate Anthropic item is titled as the launch of an opt-in vulnerability-finding service for open-source software.

The announcement is candid about why now. Anthropic writes that frontier models can be misused to exploit vulnerabilities and conduct cyber operations, and that state-sponsored adversaries have spent years gaining footholds in these systems. It says defenders of critical infrastructure and the open-source community have decades of experience but face severe resource shortages. On Project Glasswing, it says partners uncovered many vulnerabilities but that it has not yet achieved a sufficient reduction in cyber risk. In its words, finding vulnerabilities is easier than ever, while verifying, prioritizing and fixing them remains hard. Earlier in the week, Anthropic merged Glasswing into an expanded Cyber Verification Program, which gives many more defenders access to its most capable models. It says it will also provide engineering talent, tools and funding.

Sources: Introducing the Anthropic Cyber Mission — Anthropic · OSS Scanner — Anthropic (Frontier Red Team) · Launching an opt-in vulnerability-finding service for open-source software — Anthropic

A Usage Policy rewritten for longer, more independent work

Also on October 8, Anthropic published its annual Usage Policy update, which takes effect on November 12. It says most changes clarify existing rules and that the new examples reflect how Claude has taken on longer, more independent work over the past year. The update draws on misuse patterns in influence operations, weapons development and surveillance that Anthropic documented in its latest threat intelligence report.

The most visible structural change is a new section, Do Not Engage in Deceptive Campaigns or Artificial Activity. Anthropic says its policy already prohibited such activity, but the rules were scattered across the elections, fraud, privacy and disinformation sections. The new section consolidates them and covers deceptive activity of any kind, political or commercial. That includes obscuring who is behind a message, amplifying content through fake accounts, and building tools and infrastructure for influence operations. The elections section is now narrower, focused on using Claude to deceive voters or disrupt elections. Anthropic also says it clarified requirements for high-risk uses in areas such as health and finance, added controls for when Claude autonomously takes physical actions, and addressed abusive behavior toward its models.

Sources: 2026 Usage Policy update — Anthropic

Research headlines on replication, distillation and unintended actions

Three Anthropic items in today's feed concern the behavior and protection of models. One is titled 'AI models may be capable of self-replicating without access to a datacenter'. Another is 'Strengthening our defenses against distillation attacks'. The third is 'Investigating unintended model actions in our evaluations and internal use'. The material available for this edition gives their titles but not their findings, methods or stated causes. We therefore do not characterize their content beyond the headlines. Readers should go to the originals for the substance, in particular for how Anthropic explains the circumstances of the unintended actions it investigated.

A fourth item points to a different use of the models. Anthropic presents a project that uses Claude Science to produce what it titles the first complete map of the sky in UV light. Again, only the headline is available to us.

Sources: AI models may be capable of self-replicating without access to a datacenter — Anthropic · Strengthening our defenses against distillation attacks — Anthropic · Investigating unintended model actions in our evaluations and internal use — Anthropic · Using Claude Science to produce the first complete map of the sky in UV light — Anthropic

Hooks that fail closed, and agents that orchestrate agents

Three Claude Code releases landed in quick succession. Version 2.1.294 fixed prompt and agent hooks written as instructions, such as 'Block commands that...', which were allowing what they were meant to block. Version 2.1.295 added an onFailure option set to block for command and HTTP hooks. A hook that cannot start, times out or exits with an unexpected code then blocks the action instead of letting it through. Read together, the two changes make hooks more dependable as a control point. The same release added support for the Program Status Protocol (OSC 7501), so compatible terminals can show whether Claude Code is working, waiting or done. It also made the /copy picker copy drafted messages without their quote markers.

Version 2.1.296 concerns multi-agent setups. Subagents can now auto-compact earlier than the main conversation through an autoCompactWindow setting in frontmatter and --agents definitions. A new CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL variable runs every workflow agent on one model while other subagents keep theirs. The Claude apps gateway's managed.policies[] also gains a code key. It carries the same settings as cli, applies them in Claude Desktop's Code tab, and, beside desktop, turns on Claude Desktop's gateway mode. On the API side, the October 9 release notes say a Claude Managed Agents agent can now use dynamic workflows, in beta behind the managed-agents-2026-04-01 header. For work with many pieces, such as reviewing hundreds of documents, the agent can write a workflow, which Anthropic defines as a program that runs many agents in phases.

Sources: Claude Code 2.1.296 release notes — Claude Code (GitHub) · Claude Code 2.1.295 release notes — Claude Code (GitHub) · Claude Code 2.1.294 release notes — Claude Code (GitHub) · Claude Managed Agents dynamic workflows (API release notes, October 9, 2026) — Anthropic (release notes)

This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →

Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.