Salt Security debuts Salt Code to police AI coding assistants

As AI coding assistants cross the line from novelty to default toolchain, security teams are struggling with the fact that none of these models were trained on their specific policies or regulatory obligations. Salt Code is an early attempt to bolt a policy‑enforcement layer directly into the generation loop, rather than catching insecure code later with SAST/DAST scans and human review. In practical terms, that shifts part of the governance burden from developers and security engineers to an agentic system sitting between the model and the repo.

For the AGI race, this is another example of vendors building “control planes” around increasingly autonomous systems. If large enterprises can standardize on assistants while still satisfying regulatory and internal security requirements, adoption of agentic coding will accelerate, expanding the surface area where more capable models can be deployed. It also hints at a future where every powerful model is wrapped in a mesh of guardrail services—policy engines, logging layers, evaluation agents—that collectively matter as much as the raw model. Those secondary layers may quietly become some of the most important infrastructure in an AGI world.