Consequential AI Workflow Assessment
A map of where consequential AI acts today.
Glacis starts where AI already acts and maps the path from the intended rule through model calls, tool use, permissions, control coverage, escalation paths, and the evidence a reviewer will need.
We map the agent surface, prioritize the controls, and build the evidence plan.
What we map
The agent surface.
- Model calls and agent boundaries
- Tool-call exposure and delegated authority
- Credential and data-access paths
- Prompt-injection and tool-misuse risk
- Runtime control gaps and evidence gaps
What you get
A practical hardening plan.
- Runtime control recommendations
- Evidence path for signed records
- Security-review readiness findings
- Customer-facing review artifact
- Secondary ISO 42001 / NIST AI RMF mapping where useful
How it works
A single scoped assessment.
- Scope a named AI workflow
- Review agent, tool, data, and credential boundaries
- Map control and evidence gaps
- Demonstrate the record-verification and review-packet pattern
- Leave with a customer-ready evidence plan
Assessment output
A risky workflow becomes a security-review artifact.
The assessment is built for AI-native teams selling into enterprise. It focuses the conversation on what the workflow can do, which runtime controls should run, what evidence is missing, and which records your customer can inspect. Ongoing runtime coverage is priced separately, scoped to the deployment.
Surface
Risk
Control plan
Evidence path
Agent uses tools, credentials, customer data, code, or production systems.
Prompt injection, tool misuse, data leakage, unauthorized action, and drift.
Allow, block, redact, restrict, escalate, or require review at runtime.
Signed records assembled into a customer-facing evidence pack.