Discussion

GitHub’s Autofix advice: clear the risky alerts, not just the backlog

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#12811

GitHub’s Marcelo Oliveira says its Agentic Autofix tool can rerun security checks and project tests on proposed fixes, while urging security teams to prioritise risk rather than turn an old alert backlog into a mountain of pull requests. His advice is useful well beyond GitHub’s own tools: more automated fixes make review and measurement matter more, not less.

Watch Desk analysis

What happened

In an interview with Help Net Security, Oliveira, GitHub’s vice-president of product management, discusses accountability when an AI-generated patch fixes one weakness but introduces another. He says GitHub is accountable for the quality of fixes it suggests, while customers decide what gets merged.

Oliveira says Agentic Autofix can rerun detection engines and project tests to check proposed fixes. He also describes using campaigns to group fixes and let customers control their pace, rather than having an agent flood a team with as many as 2,000 pull requests overnight.

Why it matters

Oliveira recommends validating and deduplicating historical findings, then prioritising them by evidence of exploitation, whether the affected code is reachable in the deployed application, and business criticality. That gives teams a practical starting point: a large alert count is not, by itself, a sensible queue.

For measuring whether security is improving, he favours tracking high-risk vulnerability exposure over time and whether fewer vulnerabilities are being introduced or reaching production, rather than relying only on mean time to remediate. He says teams should keep scanning scope and coverage consistent, and distinguish verified fixes from alerts that were suppressed or reclassified.

Our read

The useful advice is to make automation answer to risk and evidence, not to the satisfying but sometimes misleading sight of a shrinking alert counter. Rerunning tests and detection engines is a sensible check, but it does not make human review optional. GitHub’s claims about its tools and results are claims from a company executive, not independent measurements of performance.

What to watch

  • Whether teams using Autofix publish results on verified fixes and regressions.
  • Whether vulnerability exposure falls without scanning coverage changing.
  • How GitHub documents model selection and evaluates new models.

Discussion spark: Should security teams judge AI autofix tools mainly by how quickly they close alerts, or by whether they reduce verified high-risk exposure without creating new vulnerabilities?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.