Erin Havens

Havens studied computer science and art history at McGill and wrote an undergraduate thesis on Frank Stella’s Black Paintings. Her career moved through Visa security products and GitHub’s Dependabot alerts before secret scanning. Her professional portfolio describes leading a purpose-built detection model from research to launch with Microsoft Applied Sciences.[1]

Stop a credential before it becomes public history

In her secret-protection article, Havens separates detection before a push from remediation after exposure. A blocked push can be corrected before the credential enters repository history. Afterward, revocation and replacement require coordination. Her analysis of nine quarters of pushes distinguishes rising total activity from the per-push prevalence of detected secrets.[2]

Use context within a fast blocking decision

The article describes a ModernBERT classifier that assesses candidate secrets in surrounding code. It targets values that lack a recognizable provider-specific pattern, such as a password inside a configuration file. Precision, speed, throughput and cost all matter because a false block interrupts a developer. The model is designed to classify candidates, rather than generate an explanation in the push path.[2]

Reduce repetitive alert decisions

Her Dependabot auto-triage introduction describes rules that dismiss alerts meeting selected conditions. A policy can encode routine decisions so developers spend attention on the alerts that remain. This earlier product work addresses a related security problem: detection becomes more useful when the surrounding workflow helps someone decide and act.[3]