HAIEC makes deterministic AI application source security available to developers

In August 2026, HAIEC published AI AppSec, an open-source developer security tool focused on AI applications and agent source code. It uses a deterministic static analysis approach to surface source-level findings that teams can evaluate before merging or deploying code.

The documented initial release bundles 122 detectors covering 79 security checks, using Semgrep as its analysis engine. Teams can run the tool against repositories, preserve structured findings, and integrate source checks into CI or MCP-compatible development workflows. Once configured, normal scans can run offline.

AI AppSec addresses source-code evidence. It does not establish runtime behavior, tenant boundary correctness across all paths, or human delegation of consequential choices. HAIEC positions these as separate evidence and control questions so source findings can become part of a wider assurance record without being overstated.

AI applications create new pathways from untrusted input to tools, model context, business actions and external data. General static analyzers remain essential, but agent integrations also benefit from detection patterns specific to prompt injection sinks, tool invocation logic and insecure data handling.

AI AppSec responds to that need with a public, Semgrep-backed security rulepack for AI application code. Its checks can surface known risky implementation patterns before a change is merged or deployed, making the review more repeatable across developers and repositories.

The documented initial release lists 122 detectors across 79 checks. Its output is intended for developers and security reviewers, not simply for creating a headline count. A useful result names the source location, relevant rule and context that makes the pattern worth investigating.

Teams can integrate analysis into selected developer workflows and carry structured results into review. The tool is suited to targeted source-level investigation; a clean report does not prove that model outputs, effective IAM grants or live actions are safe.

HAIEC maintains complementary controls for MCP tenant isolation and runtime model input and output. Together they address different parts of the AI application attack surface while preserving the boundaries of each method.

This open-source engineering approach supports early detection and technical transparency. Enterprise assurance builds on it by asking which consequential actions are code-capable, authorized, delegated and actually evidenced within the evaluated system.