Your teams are shipping on Amazon Bedrock faster than anyone can review it. Cranium ONE finds every Bedrock model in your code, inspects every prompt and response, and hands you the evidence that it’s under control.
CodeSensor builds an AI Bill of Materials for every repository — models, datasets, technologies and infrastructure — including Bedrock Runtime calls made from .NET. AgentSensor maps agents built with Amazon Strands: their models, tools, handoffs and MCP servers.
Guardian inspects the traffic between your application and the Bedrock model. In Listen mode it evaluates a copy in parallel, so you get sessions, events and every signal triggered without touching the request path.
Profiles decide which signals matter — PII, PHI and PCI, secrets, jailbreaks, instruction override, prompt leaking — and what happens to each: block, modify or pass.
Switch a profile to Enforce and Guardian sits between your app and Bedrock, acting on prompts and responses before they land. Prove the profile first with Cannon, which runs whole prompt collections through it in one run.
Trace explains every verdict from deterministic, purpose-built classifiers, and AI Cards turn the record into evidence mapped to the EU AI Act, NIST AI RMF and ISO 42001.
Cranium ONE works with the stack you already run. Your policy stays yours — we make it enforceable.
In Guardian, add a Bedrock model with its vendor model ID and the IAM Role ARN Cranium should assume.
Point your application through Guardian. Listen mode evaluates traffic in parallel and changes nothing.
Tune the profile, test it with Cannon, then switch to Enforce when the verdicts look right.
You add a Bedrock model in Guardian with its vendor model ID and an IAM Role ARN. Cranium uses that role to reach the model, so there are no long-lived access keys to store.
Only if you choose. In Listen mode Guardian evaluates a copy of the traffic in parallel and changes nothing. In Enforce mode it runs in line between your application and the model, and can block, modify or pass each prompt and response.
CodeSensor scans your GitHub, GitLab, Bitbucket and Azure DevOps repositories and records the models each one calls in an AI Bill of Materials, including Bedrock Runtime calls from .NET code. AgentSensor maps agents built with Amazon Strands. Scans can run weekly, monthly or on every push.
The signals your profile enables, including PII, PHI and PCI, secrets, jailbreaking, instruction override, prompt leaking, role impersonation and toxicity. Every verdict comes from deterministic, purpose-built classifiers and is explained in Trace.
We use essential cookies to run this site, and — only with your permission — analytics and marketing cookies to improve it. You're the hero here; you stay in control. Privacy Policy.
OpenAI
Anthropic Claude
Microsoft Azure