Anthropic launches a Cyber Mission for defenders and rewrites its Usage Policy

A Cyber Mission aimed at defenders, with open source as the first deliverable

On October 8, Anthropic announced the Anthropic Cyber Mission, described as a long-term commitment to securing the systems everyone depends on. It starts with critical infrastructure, namely the operational technology behind power grids, water systems and transportation networks, plus government systems. It also covers open-source software. The Critical Infrastructure Defense Program brings frontier models, on-site engineers and threat research to the people who defend that technology. OSS Scanner offers open-source projects regular security scans from Anthropic's strongest models, at no cost. Anthropic presents it as an opt-in vulnerability-finding service.

The announcement states its reasoning openly. Anthropic says 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 also says defenders of critical infrastructure and open source have deep experience but severe resource shortages. On Project Glasswing, Anthropic says partners uncovered many vulnerabilities but that it has not yet achieved a sufficient reduction in cyber risk. Finding flaws is now easier than ever, it says, while verifying, prioritizing and fixing them remains hard. Earlier in the week it merged Glasswing into an expanded Cyber Verification Program, which gives many more defenders access to its most capable models.

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

A Usage Policy that follows Claude into longer, more independent work

Anthropic published its 2026 Usage Policy update on October 8. The new version takes effect on November 12. Anthropic describes most of the changes as clarifications of existing rules. It says that over the past year Claude has taken on longer, more independent work, so the update adds examples of how the rules apply to those capabilities. It also draws on new patterns of misuse in influence operations, weapons development and surveillance, which Anthropic documented in its latest threat intelligence report. The update clarifies requirements for high-risk uses such as health and finance. It adds controls for cases where Claude autonomously takes physical actions, and it addresses abusive behavior toward the models.

The main structural change is a new section, “Do Not Engage in Deceptive Campaigns or Artificial Activity.” Anthropic says it has seen state media outlets, government propaganda offices and commercial firms use Claude to run networks of fake accounts and fabricated news sites. The company says its policy already prohibited this, but the rules were scattered across the elections, fraud, privacy and disinformation sections. They are now consolidated and apply to deceptive activity of any kind, political or commercial. The elections section is refocused on barring the use of Claude to deceive voters or disrupt elections. Anthropic also published a post titled “Investigating unintended model actions in our evaluations and internal use.” The material available to us carries only that title, so we do not describe its content.

Sources: 2026 Usage Policy update — Anthropic · Investigating unintended model actions in our evaluations and internal use — Anthropic

Agent tooling: workflows that scale up, and hooks that fail closed

The October 9 API release notes say a Claude Managed Agents agent can now use dynamic workflows. This is 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 is a program that runs many agents in phases. Claude Code 2.1.296 fits the same direction. A subagent can now auto-compact earlier than the main conversation through the new autoCompactWindow setting. The new CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL variable runs every workflow agent on a single model while other subagents keep theirs. The release also adds a code key to the managed.policies[] of the Claude apps gateway. It carries the same settings as cli and is also applied in Claude Desktop's Code tab. Used alongside desktop, it turns on Claude Desktop's gateway mode.

The two preceding releases concern hook reliability. Version 2.1.295 adds onFailure: “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. The same release supports the Program Status Protocol (OSC 7501), so compatible terminals can show whether Claude Code is working, waiting or done. It also adds quoted text to the /copy picker, so a drafted message copies without its “>” markers. Version 2.1.294 fixed prompt and agent hooks written as instructions, such as “Block commands that...”, which were allowing what they should have blocked. It also improved how Stop and SubagentStop prompt hooks written as instructions are judged, so Claude is less likely to stop early.

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

Science on the publication list

Anthropic also published two items on science. One is titled “Using Claude Science to produce the first complete map of the sky in UV light.” The other is “Building on our commitment to American scientific discovery.” The material available to us carries only the titles, so we do not report details of either.

Sources: Using Claude Science to produce the first complete map of the sky in UV light — Anthropic · Building on our commitment to American scientific discovery — Anthropic