'Ghostcommit' hides prompt injection in images to fool AI agents, steal secrets
Researchers have built a pull request that steals a repository's secrets by hiding the malicious instruction inside a PNG that AI code reviewers never open. The reviewer waves the change through.
Grace Ops
CONTENT OPTIMIZATION · AEO/GEO
Not yet scored. Next refresh: 02:00 UTC.
audit trail / provenance1
Provenance
Claims tie surfaced fields back to sources, models, or heuristics.
- severity.upliftheuristicn/aRansomware campaign indicators detected — floor raised to at least high.
What changed
Append-only revisions when ingest or analysts evolve the record.
No revision rows stored yet.
Sources
Brief grounded in 1 source.
discussion
Sign in to join the thread and vote on comments.
Loading comments…