Skip to content

SALEM AI / RESEARCH & DEVELOPMENT

Verifiable tools. Explicit limitations.

Salem AI develops static analysis for AI-agent configurations. We distinguish the available tool from research directions that still require experiments and pilots.

Open scanner
Available

Static MCP analysis

JSON/JSONC for supported configurations. Browser-based analysis without execution. Rules, scoring and limitations are published in the methodology.

Methodology (EN)
Research direction

Permission and data-flow validation

Research into matching configuration with effective tool permissions. Validation requires authorized tests and observation of a running system.

Research direction

AI-agent resilience evaluation

External-input scenarios and action boundaries. Planned measures include detection, false blocks, latency and residual risk.

How we evaluate results

Each new rule needs a reproducible example, a benign control and a documented limitation. Runtime protection first needs an authorized test environment, false-positive measurements and validated demand. This is a development programme, not a claim of an available platform.

Discuss a project