---
type: "post"
id: "github-brings-its-fine-tuned-secret-detection-model-03c54b"
slug: "github-brings-its-fine-tuned-secret-detection-model-03c54b"
title: "GitHub brings its fine-tuned secret-detection model to push protection and Copilot /security-review"
description: "GitHub has moved its fine-tuned leaked-secret detection model into more parts of its security workflow: existing secret alerts, private-preview push protection, and forthcoming Copilot security-review checks."
retrieval_nugget: "GitHub has moved its fine-tuned leaked-secret detection model into more parts of its security workflow: existing secret alerts, private-preview push protection, and forthcoming Copilot security-review checks. For AI-building teams, the practical change is not only better detection but a new opt-in, credit-metered control point before code reaches repository history."
published_at: "2026-10-28"
updated_at: "2026-10-28"
record_date: "2026-10-28"
date_kind: "published_at"
topics: ["github","code-review"]
entities: ["GitHub","GitHub Secret Protection","GitHub Advanced Security","GitHub Copilot","Copilot CLI","GitHub Enterprise Server"]
related_patterns: ["verification-bandwidth-is-the-scarce-resource"]
novelty_label: "new_fact"
what_changed: "GitHub is extending its fine-tuned secret-detection model to push protection and Copilot security reviews, with preview timing and AI Credits billing details."
prior_refs: ["memory:946bb361-44bc-8caa-a65c-ffc20eb20938","memory:08b471cb-3d83-8a92-a14e-c0f7e6663cb9"]
source_url: "https://github.blog/changelog/2026-10-07-purpose-built-model-for-leaked-secret-detection"
source_urls: ["https://github.blog/changelog/2026-10-07-purpose-built-model-for-leaked-secret-detection"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:github-brings-its-fine-tuned-secret-detection-model-03c54b"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/github-brings-its-fine-tuned-secret-detection-model-03c54b/","markdown":"https://newruntime.com/posts/github-brings-its-fine-tuned-secret-detection-model-03c54b.md","json":"https://newruntime.com/posts/github-brings-its-fine-tuned-secret-detection-model-03c54b.json"}
---

# GitHub brings its fine-tuned secret-detection model to push protection and Copilot /security-review

## Retrieval answer

GitHub has moved its fine-tuned leaked-secret detection model into more parts of its security workflow: existing secret alerts, private-preview push protection, and forthcoming Copilot security-review checks. For AI-building teams, the practical change is not only better detection but a new opt-in, credit-metered control point before code reaches repository history.

## Summary
GitHub has moved its fine-tuned leaked-secret detection model into more parts of its security workflow: existing secret alerts, private-preview push protection, and forthcoming Copilot security-review checks. For AI-building teams, the practical change is not only better detection but a new opt-in, credit-metered control point before code reaches repository history.

## What happened
On October 7, 2026, GitHub announced plans to use its purpose-built secret-detection model across secret scanning alerts, push protection, and GitHub Copilot security reviews. The model reads surrounding code to identify likely credentials, including passwords that do not have a recognizable token format. GitHub says it does this without generating code or prose.

Existing customers using AI-detected Password alerts have already been automatically moved to the new model. Those alert scans remain included, at no additional charge, for customers with GitHub Secret Protection (GHSP) or GitHub Advanced Security (GHAS). GitHub also plans to bring AI-detected alerts to GitHub Enterprise Server 3.23 in public preview, included with an enterprise's existing GHSP and GHAS purchase.

The more consequential new workflow is AI secret detection in push protection. It is available in private preview for customers on GitHub Enterprise Cloud or GitHub Team that have paid GHSP or GHAS coverage. The check examines unstructured credentials at push time, so a developer can remove a likely secret before it enters repository history. An administrator must enable it under the applicable organization or enterprise policies. GitHub says the planned AI Credit usage is billed to the organization that owns the repository, except in specified enterprise-managed-user repositories where usage is attributed to the pusher. A check can consume credits even when it does not block a push.

GitHub is also adding secret-classifier checks to the `/security-review` command in supported Copilot CLI and Copilot App sessions. The command reviews active changes for vulnerabilities, returns prioritized findings and remediation suggestions, and can be run again after confirmed findings are addressed. The review is read-only, while existing Copilot policies and billing still apply. The classifier checks are coming soon in private preview, are off by default, and require a separate opt-in; merely running `/security-review` does not activate them. They do not require a GHSP or GHAS license, but they will consume AI Credits in addition to the review's existing usage and be reported under GHSP in AI usage insights.

GitHub distinguishes hosting plans, security licenses, and Copilot subscriptions. Individual Copilot plans, including Free and Student, can be eligible for the security-review checks subject to access controls and credit consumption. Copilot Business and Enterprise plans can also be eligible on supported platforms, but they do not replace the GHSP license required for AI push protection. Administrators can disable the new capabilities by policy and set AI Credit budgets. GitHub notes that budget alerts alone do not stop usage; a stop-usage limit must be configured where available.

## What the archive adds
This is a new fact in an existing GitHub security story. The October secret-scanning update covered encoded-token revocation, API and webhook metadata fixes, and an audit-log migration in [Updates to GitHub secret scanning](https://newruntime.com/posts/updates-to-github-secret-scanning-8b5197/). This announcement instead introduces a fine-tuned classifier, new workflow surfaces, and AI Credit billing. It also follows [GitHub Copilot code review Balanced depth is generally available](https://newruntime.com/posts/github-copilot-code-review-balanced-depth-is-generally-available/), which documented a separate control over review depth.

## Trend signal
This modestly supports [Verification bandwidth is the scarce engineering resource](https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource/). GitHub is not claiming that its model removes human security judgment: it places another classifier in a push-time gate and in a review loop that returns prioritized findings, supports remediation, and expects a rerun. The workflow is therefore organized around finding and checking risky changes earlier rather than around unconstrained code generation.

## Why now
GitHub frames the rollout around software built both directly by developers and with AI agents. The timing also coincides with a billing boundary: included AI-detected alerts are separated from new, opt-in checks that consume AI Credits. Publishing the pricing and attribution rules before broader availability gives organizations a chance to set policies and budgets before enabling those checks.

## Connections to other events
GitHub had already made Balanced depth generally available for Copilot code review, giving teams an explicit review-depth control. This announcement adds a different kind of specialization: a secret classifier alongside the existing LLM-based security review. Separately, an earlier archive item described a security-review workbench that can scan repositories, paths, diffs, and working trees. The present change is narrower: it adds secret-focused checks to GitHub's own Copilot review command and links them to a defined credit and policy model.

## Why it matters
A leaked credential is unusually expensive to discover after it has landed in repository history, propagated into a pull request, or reached an external service. GitHub's push-time option makes secret detection a preventative control rather than solely an alerting function. The Copilot integration also gives teams a way to put secret checks beside vulnerability review before commit, push, or pull-request review.

The operational trade-off is explicit. Teams now need to decide which checks belong in developer and agent workflows, who may enable them, which billing account pays, and whether budget limits actually stop consumption. That is useful discipline for AI-assisted security tooling: a detection feature is no longer just a model capability when it can run automatically at high-frequency development boundaries.

## What happens next
- **GitHub will broaden access to the opt-in push-protection checks after private preview.** (3 months, 65%). Watch for: a public-preview or general-availability notice with final credit rates.
- **Organizations will add budget and policy controls before enabling checks widely.** (6 months, 70%). Watch for: documentation or product controls focused on credit caps and organization-level rollout.
- **Secret detection will become a more visible component of Copilot security review.** (6 months, 60%). Watch for: supported-version requirements and classifier findings in the review output.

## Limits and uncertainty
GitHub's announcement describes availability states and planned billing, not measured effectiveness. It does not provide false-positive rates, recall, comparisons with existing secret scanning, typical credit consumption per check, or evidence that the new model improves outcomes in particular languages or repositories. Push protection is only in private preview, and the Copilot secret checks are described as coming soon in private preview, so neither should be treated as generally available. The source also does not establish how organizations will tune findings, resolve disputed detections, or evaluate the security value against credit spend.

## Skill to build now
Map the points where credentials can enter your repositories: local changes, agent-authored edits, commits, pushes, and pull requests. Then define an owner for each proposed AI security check, the authorization needed to remediate a confirmed finding, and a budget that can halt usage rather than merely alert on it. Run a small trial with representative repositories and record whether findings are actionable, whether developers rerun review after fixes, and how often checks run without blocking a push.

## Reader takeaway
GitHub is turning secret detection from an included scanning feature into a set of workflow controls with different availability, permissions, and costs. The useful question for builders is not simply whether to enable the model: it is where secret checks should run, who can activate them, and how to prevent their automated use from becoming unbounded spend.

## Open question
Will the fine-tuned classifier provide enough measurable improvement over existing secret scanning to justify credit-consuming checks at push and review time?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Github](https://newruntime.com/topics/github.md): Explore the github topic hub.
- topic hub: [Code Review](https://newruntime.com/topics/code-review.md): Explore the code-review topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)