Multi-model access hides multi-model risk
Bedrock makes it trivial to call five different foundation models from one API, but each model call is still a distinct trust decision.
By usage
Every model and agent on Bedrock, governed at the gateway.
Bedrock makes it trivial to call Claude, Titan, Llama, and Nova from one API, and to build Bedrock Agents with action groups and knowledge bases. IAM scopes which APIs a role can call - it says nothing about what happens once the call goes through.
Why it matters
Multi-model access hides multi-model risk
IAM scopes access, not behavior
An IAM role can limit which Bedrock APIs a service can call, but it says nothing about what a model does with the tokens once invoked.
Bedrock Agents add tool-calling risk
Agents with action groups and knowledge bases can reach internal APIs and data stores - the same MCP-style risk surface as any other autonomous agent.
What Lineation does
Policy and lineage on top of IAM
Cross-model governance in one layer
Claude, Titan, Llama, and Nova calls through Bedrock are governed under one policy model instead of per-model configuration.
IAM-aligned agent identity
Bedrock Agents and service roles map to scoped Lineation identities, extending AWS IAM boundaries with policy at the tool-call level.
Action-group & knowledge-base policy
Tool calls a Bedrock Agent makes through action groups are policy-checked before they execute.
Full request/response lineage
Every invocation across every model is logged and replayable for audit and cost review.
How it works
Five steps to a governed Bedrock deployment
Route Bedrock traffic through the gateway
Model invocations and Bedrock Agent tool calls pass through Lineation's LLM/MCP gateway.
Map IAM roles to agent identity
Service roles and Bedrock Agents resolve into scoped identities in the inventory.
Apply cross-model policy
Default-deny rules apply consistently across whichever foundation model is invoked.
Govern action groups
Tool calls from Bedrock Agents are checked against policy before reaching internal APIs or data stores.
Review lineage & spend
Every invocation is logged and replayable, with per-model and per-agent cost attribution.
FAQ
Common questions
Which Bedrock models does this cover?
Any model invoked through Bedrock - Claude, Titan, Llama, Nova, and others - is governed under the same policy layer.
How does this relate to AWS IAM?
IAM continues to scope which APIs a role can call; Lineation adds a policy and lineage layer on top that governs what happens inside those calls.
Does this cover Bedrock Agents and Knowledge Bases?
Yes - action-group tool calls and knowledge-base retrievals are policy-checked the same way any other agent tool call is.
Do we need to change how we invoke Bedrock today?
No. Traffic routes through the gateway at the network layer; application code calling Bedrock APIs is unchanged.
Can we track spend per model and per agent?
Yes - invocation-level lineage supports per-model and per-agent cost attribution, the same data used for token spend control.
Does this work alongside AWS-native monitoring like CloudTrail?
Yes - Lineation complements CloudTrail with policy enforcement and content-level inspection CloudTrail doesn't provide.
Govern every model on Bedrock, from one layer.
Connect your Bedrock traffic and see cross-model activity today.