Accidentally commit a live access token to a repository and you have two jobs. One is to make sure nobody can use that credential. The other is to fix the code that exposed it. Deleting the offending line might feel productive, but anyone who already copied the token still has a working credential.
GitLab 19.6 brings those jobs together with a feature called “Rotate and fix,” now in beta on GitLab.com. When GitLab detects an active GitLab personal access token in code, a team can start the cleanup in one step. GitLab revokes the exposed token, creates a replacement with the same permissions, and stores it in GitLab Secrets Manager.
So where does the AI come in? GitLab Duo opens a merge request to remove the hardcoded token. By the time a developer sits down to review the proposed change, the old credential has already been revoked. That is the part I like best. Nobody has to wait for an AI-generated patch to get approved before the exposed token stops working.
There are a couple of details I’d want to check as a developer. The replacement inherits the permissions of the original, so an overly generous token comes back with the same privileges. And if another job depends on the old token, it will need access to the new one. GitLab describes how it stores the replacement and proposes the code change. It makes no claim that it finds every job using that credential.
Another beta feature tackles a question familiar to anyone who has ever looked at a long list of security findings. Which one do you fix first? GitLab’s new vulnerability context flow examines findings from static application security testing and helps teams assess whether a flaw is exploitable in practice. A severity rating tells you how bad a vulnerability could be. Knowing whether someone could actually reach the affected code gives a developer a better reason to move one fix ahead of another.
Teams can also run security agent flows on branches other than the default branch, giving developers a chance to address a finding before their code merges. At the group level, an organization can require a designated person to approve changes involving critical or known-exploited vulnerabilities.
GitLab is also putting its Duo agents on a schedule. The beta lets teams run recurring tasks as often as every 15 minutes or as infrequently as every three months. A team might have an agent check its continuous integration jobs every hour and post a likely cause when a build fails. Each scheduled automation gets its own service account with limited permissions, so someone looking through the audit records can tell which changes came from the agent.
For developers who spend much of the day in Slack, there’s a more immediate way to put Duo to work. Premium and Ultimate customers on GitLab.com can now mention @GitLab in a conversation and ask it to turn the discussion into an issue or a merge request. The request uses that person’s GitLab permissions and consumes GitLab Credits. If the agent proposes a code change, the usual review and approval rules still apply.
As for the cost, GitLab says it has reduced the number of model requests Duo needs to complete a task. It expects most customers to get up to twice as much agent work from their credits. New customers get the change November 19, while existing customers move over December 18 unless they opt in sooner.
I’d like to see what the same job costs before and after that change, particularly for something a team plans to run every hour. In the meantime, “Rotate and fix” gives developers a more concrete reason to look at GitLab 19.6. If I accidentally publish a token, the first thing I want to know is whether it still works. GitLab can take care of that while I review the patch.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz


