Droid Shield 2.0

Factory's Droid Shield 2.0 adds contextual judgment to scanning code for exposed access credentials. A pattern scanner can mistake an example value for a real secret or miss credentials outside its known patterns. The research preview addresses these errors with two separately trained classifiers.

One examines suspicious lines the scanner missed. The other reviews scanner alerts and decides whether they are false alarms. Only the second receives a window with detected credential values masked; the first sees the candidate value. Adjustable thresholds govern when to warn or clear an alert, making this a configurable balance between missed secrets and unnecessary blocks.

“Droid Shield is an extra line of defense against exposing potential secrets during an autonomous commit and push.”

“We are releasing the model weights in the interest of compounding future work on software security and privacy.”

“The model must decide from context alone whether the scanner hit should stay blocked or be cleared as a false alarm.”

“We mask every known credential span across the entire window, so the model learns to judge without ever seeing the secret value(s).”