Anthropic ships a vulnerability scanner into Claude Code itself — Claude Security enters beta

Anthropic launched Claude Security in beta on 22 July 2026, a multi-agent vulnerability scanner built into Claude Code that reviews uncommitted changes or a full repository from the terminal and proposes patches a human still has to approve — a direct product answer to the AI-code-trust-gap story this site has tracked all month.

Anthropic released Claude Security as an official Claude Code plugin in beta on 22 July 2026. It adds a single command, /claude-security, with two headline jobs: scan your uncommitted changes — a branch’s diff, a pull request’s diff, or a single commit — before you commit, or run a full scan across the whole repository, or a scoped subset of it. Both run from the terminal, on the same Claude inference developers are already paying for through their Claude Code plan.

The mechanism is a multi-agent pipeline, not a single linear pass. One agent maps the code’s architecture first, so the scan has context on how the pieces connect. A second identifies candidate vulnerabilities against that map. A third verifies each finding specifically to cut false positives — the complaint that has dogged every automated security scanner since static analysis existed. A fourth proposes a style-matched patch for anything confirmed. Nothing is auto-applied: a human reviews and approves every fix before it lands, and Anthropic is explicit that the tool never auto-commits changes on its own. The one practical caveat surfacing in early coverage is cost — a full-repository scan on a large codebase can burn meaningfully more tokens than a scoped diff scan, and may shift work onto pricier models, so teams are advised to default to scoped runs and reserve full scans for periodic sweeps rather than every commit.

So what

This is the vendor half of a story this site has been building all month from the other direction. The trust-gap posts — Stack Overflow’s confidence numbers, the AI Now Institute’s Friendly Fire paper, Check Point’s live-attack findings, the 12-fix permission-hardening sprint covered here on 23 July — have all documented the same underlying tension: AI coding agents write a growing share of shipped code, and the tooling to verify what they wrote hasn’t kept pace with the tooling to generate it. Claude Security is Anthropic’s attempt to close that gap with the same agent doing the writing, rather than routing everything through a separate SAST vendor bolted on after the fact. Whether a Claude-native scanner catches what an independent tool would miss is an open question worth watching rather than taking on faith — verify-the-verifier is still good practice. But for teams already committed to Claude Code, a scoped-diff scan before every commit is a close-to-free addition to the workflow, and it’s the kind of governance layer we build into every AI-assisted development engagement regardless of which vendor’s scanner sits behind it. If you’re weighing how to structure review and approval around AI-generated code on a live build, get in touch.

← All research

More research.

GitHub secret scanning cuts false alarms for AWS, Google and Alibaba cloud keys and adds a SpaceX detector

GitHub's 5 November 2026 secret scanning update tightens detectors for AWS, Google Cloud and Alibaba keys, relevant to teams building software with AI coding tools that can paste credentials into code.

Read ›

GitHub's Copilot code review can now be managed by API, with review effort levels and AI credit costs shown

GitHub's 6 November 2026 update lets organisations manage Copilot code review by API and see the review effort and AI credits each AI code review used, a sign AI-assisted development is becoming auditable.

Read ›

GitHub's Copilot impact dashboard now counts chat and VS Code agent use, and drops the 'passive users' label

On 6 November 2026 GitHub added Copilot Chat and the VS Code Agents window to its Copilot impact dashboard, bringing it to nine tracked experiences, and renamed 'Passive users' to 'Chat-first users'.

Read ›

Claude Code 2.1.319: a setting that denies permission prompts so AI agent runs don't stall

Claude Code 2.1.319 (6 November 2026) adds a permissions.denyInsteadOfPrompting setting and moves long MCP tool calls to the background, a sign AI-assisted development is being built for unattended runs.

Read ›

GitHub now reports how often developers act on Copilot code review comments

On 5 November 2026 GitHub added code review metrics, including an addressed rate, to the Copilot usage metrics API. For AI software development buyers, it is the first first-party number for how much AI review feedback is acted on.

Read ›

Dependabot can now keep AI agent skills and prompts up to date

GitHub's Dependabot gained Agent Package Manager (APM) support on 3 November 2026, treating AI agent skills, prompts and instructions as dependencies that need updating like any other software.

Read ›

Start a project

Building with AI? Let's talk.

Tell us what you are building. We'll give you an honest view of the right approach.

Start a projectSee our work