Cranium AI
An AI security and governance platform. Cranium ONE builds AI‑BOMs from source code, inspects and enforces policy on live prompts and responses, red teams models and shares the evidence through AI Cards.
Why consider it instead of Noma SecurityAdds AI‑BOM export in CycloneDX and SPDX and a request-and-review workflow for vendor AI Cards. Noma documents neither.
- AI‑BOMs from code
Scans GitHub, GitLab, Bitbucket and Azure DevOps repositories into AI‑BOMs of models, datasets, packages, agents and MCP servers, exported as CycloneDX or SPDX.7,8,9,10
- Inline enforcement
In Enforce mode, Guardian blocks, modifies or passes prompts and responses inline, for models connected through OpenAI, Anthropic Claude or AWS Bedrock.11,12
- Evidence for auditors and vendors
Maps AI systems to the EU AI Act, NIST AI RMF and ISO/IEC 42001, and AI Cards let you request, review and track AI evidence from vendors.13,14
- Best for
Teams that must secure AI and prove it: AI‑BOMs, red teaming, runtime protection and framework evidence from one platform.
- Tool calls are recorded, not authorized
Tool Inspector shows the tool calls an event made. Enforcement acts on prompts and responses between the application and the model; an MCP gateway and agent identity management are not offered.15,16
- Direct prompt attacks only at runtime
Runtime signals cover direct attacks such as jailbreaking, instruction override and direct command injection. Indirect prompt injection is a test category in Arena, not a runtime signal.17,18
- Discovery starts from code, hosted only
Agents and MCP servers are inventoried from source code, the Shadow AI view is built from connected SIEM activity, CloudSensor covers Azure only, and the platform is hosted in US, EU and AU regions.9,19,20,21
- Consider something else if
You need to allow or block individual agent tool calls, manage agent identities, discover agents on endpoints or inside SaaS platforms, or deploy on-premises. Noma Security documents all four.2,4,22
