Git author names are just text: signed commits, AI reviewers and a release record

Flat illustration of an edited hire agreement beside the original, a risk report and a warning sign

Short answer: the author name and email on a Git commit are just text that whoever makes the commit types in. A security vendor showed in April 2026 that an AI-powered pull-request reviewer, set up to trust “recognised” authors, could be fooled by a spoofed author name. The fix is not a smarter AI reviewer: it is signed commits, branch protection that requires human review, and a dated record of exactly what you released.

Last checked 12 November 2026. General information, not security advice for your specific systems. TimeBinder does not scan code, review pull requests or detect malicious changes, and a dated record is not proof that code is safe or who wrote it.

Table of Contents

What was shown

In a post dated 15 April 2026, the security company Manifold described an experiment on its own test repository. It set a well-known researcher’s name and email on a commit using two git config commands, then opened a pull request containing a malicious instruction file. A GitHub Actions workflow it had configured to approve pull requests from “recognised industry legends” (an intentionally explicit trust rule, as the company says) approved and merged it. The company reports that on a first attempt the AI flagged the pull request for manual review, and that the same pull request was approved when resubmitted. It also says a coding agent later followed the planted instructions in a test environment.

This was a vendor demonstration against a deliberately permissive setup, not a report of a real-world breach, and the company itself says not every repository using these tools is misconfigured. The structural point is still sound and old: Git lets anyone claim any author identity.

What a commit does and does not prove

  • Author and committer names, emails and dates are self-declared. They come from the machine that made the commit and can be set to anything. See are Git commit dates proof?
  • A “Co-authored-by” trailer is also just text. It can help you track AI-assisted work honestly, but it does not prove who or what wrote the code.
  • A signature proves control of a key. GitHub marks a commit “Verified” when a GPG, SSH or S/MIME signature checks out against a key registered to the account. That is a stronger statement than a name, though it does not show the key holder is who you think, that the key was not stolen, or that the code is safe.
  • Integrity is a separate job. A commit ID depends on the contents, so changing history changes the IDs. That tells you whether history was rewritten, not who wrote it or when.

What maintainers and teams can do

  1. Do not let any automated reviewer, AI or otherwise, decide trust from author names. Manifold’s advice is that AI review should flag, not merge.
  2. Require signed commits on protected branches. GitHub documents “Require signed commits” as a branch protection setting, and signing with SSH is the simplest route for most people.
  3. Require human approval for merges through branch protection or rulesets, regardless of what a bot decides. GitHub also lets you require that the latest push be approved by someone other than the pusher.
  4. Treat agent configuration files as code. Files that tell coding agents what to do (for example skill or rules files in editor folders) should get the same review as scripts, per Manifold’s recommendation.
  5. Give agents their own identity or make sure their commits carry a trailer, so you can tell human and agent changes apart later. It helps auditing but is not security on its own.
  6. Limit what automation can do. Least privilege for workflows that can approve or merge.

This is orientation, not a security review. Check GitHub’s current documentation for your plan and your setup.

A dated record of what you released

When a release is questioned (a customer, an auditor, a dispute with a contractor), signed commits show which keys signed, but the release itself is a set of files. Good practice is to keep:

  • the exact release artefacts and their SHA-256 hashes (our hash checker works in your browser; the file stays on your device),
  • the tag or commit ID they were built from, and the build log,
  • who approved the merge and the review record, exported if your platform may later change or delete it.

A TimeBind is a dated, checkable record that an exact file existed no later than a recorded time and still matches. Your first Pooled TimeBind is free; see how it works. For developers, that is a dated record of an exact release, kept outside the repository that could be rewritten. See developer timestamping, source code timestamping and how to prove source code existed before a date. If you also sign releases with Sigstore, this comparison explains how the two differ.

What a TimeBind does and does not show here

It shows that the exact files you gave it existed by the recorded time and have not changed since. It does not check code, show it is free of malware, confirm who wrote it, or replace signing, review and branch protection. A timestamp on a malicious file is a true record of a malicious file. It is one supporting layer, not a defence.

Limits

The main source is a vendor blog about its own test, so treat its figures (for example how many workflow files reference an AI action) as indicative, and read GitHub’s documentation for how its features work today. This is general information, not security advice for your systems.

Tools that can help

There is no affiliate link on this page: nothing here depends on a paid third-party tool. Our affiliate disclosure explains when we do use them.

Frequently Asked Questions

  1. Can someone fake the author of a Git commit?

    Yes. The author and committer name, email and dates are set locally by whoever makes the commit, so they can be set to anything. A verified signature is a stronger sign, because it ties the commit to a key registered to an account.

  2. Does a signed commit prove who wrote the code?

    It shows the commit was signed with a key GitHub has registered to that account. It does not prove the key holder is who you think, that the key was not stolen, or that the code is safe. Reviewers still need to review.

  3. Should an AI tool approve or merge pull requests?

    A security vendor’s April 2026 test suggests an AI reviewer should not decide trust from author names. Its advice is that AI review should flag, and that branch protection should still require human approval.

  4. Can a timestamp show my release is safe?

    No. A TimeBind records that exact files existed no later than a recorded time and still match. It does not scan code, show who wrote it or show that it is safe.

Sources checked

Keep a dated record of what you released

Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. Click here to start, or see plans and pricing.