Product Release: Threat Model

Today, depthfirst adds threat models to our platform: living security artifacts across all repositories that teams can review, tune, and apply to findings.

In our experimentation, a precise threat model has the greatest impact on the accuracy of vulnerability detection and calibration, even more than the model selection itself.

Editable threat models capture trusted inputs, untrusted paths, trust boundaries, and suggested updates for each repo component.

Security Context Is Tribal Knowledge
Modern products are made up of many components, from services, APIs, workflows, background jobs, identity systems, CI pipelines, and deployment assumptions. Security teams may understand one part of the system while Engineering teams have a stronger grasp of another, but the context needed to surface the real risk is rarely shared across the organization, let alone made available to a scanner.

That results in a cold start problem for every finding. A tool finds suspicious code, assigns a severity, and hands the result to humans to interpret. Teams then reconstruct the architecture around the finding: whether the path is reachable, an attacker controls the input, the data is sensitive, and the issue matters in the context of this application.

The Cost of Missing Context
Security is contextual. A critical issue in one application can be irrelevant in another. A bug that crosses a tenant boundary is different from one locked behind an internal-only workflow. So is a credential sitting in production infrastructure versus one hardcoded in a local test fixture.

Without a durable threat model, teams keep answering the same questions:

• Is this actually exposed?
• What trust boundary does it cross?
• What would an attacker control?
• What impact matters here?
• How do I know?

That work does not compound. The next scan forgets what the team already learned.

A Threat Model That Improves With Usage
Threat modeling has always been a valuable exercise, but maintaining accuracy and applying models effectively are difficult problems. Repositories evolve, architectures shift, attack surfaces change, and once-valid assumptions become stale. Traditional threat models rarely keep pace, as maintenance depends on manual updates from security teams that are already stretched thin.

depthfirst addresses that drift by turning tribal security knowledge into explicit, reviewable, and reusable artifacts that evolve with the codebase. It generates an initial model, then helps teams keep it alive by editing assumptions, providing suggested updates, or regenerating it with guidance.

Once saved, the model becomes durable context for vulnerability discovery and triage. Scans become stateful: accepted assumptions, severity calibration, and agent-learned signals carry forward, so each scan starts with the product’s security logic instead of relearning it from scratch.

Available Now In Private Preview
Threat Model is now available in Private Preview for depthfirst customers. If you want to see how a living threat model can improve vulnerability discovery and severity calibration for your repos, book a demo below.