Services/Artificial intelligenceNew
AI Blast Radius Assessment
A security architecture assessment for AI deployments of any size. We map what each AI system can reach, and how far the damage spreads if one is compromised.
Your AI systems are only as protected as the security architecture around them. If one part of your AI deployment is compromised, how far does the damage spread?
What we assess
Permissions and reach
What each AI system, agent and service identity can read, write and execute, and which other systems that access leads to.
Security enclave configurations
Whether the enclaves that isolate your models, data and workloads are configured the way the design says they are.
MCP servers
The Model Context Protocol servers your agents connect to: which tools and data each one exposes, and to whom.
API gateways
Authentication, authorization and routing on the gateways in front of your models and AI services.
Data loss prevention
Where relevant, whether your DLP controls cover the paths data takes into and out of your AI systems.
Governance mapping
Your controls mapped to governance frameworks such as NIST AI RMF and MITRE ATLAS.
What you receive
Blast radius report
The report details what each AI system can reach, how far a compromise of any one part would spread, and recommendations to improve the security.
Architecture findings
Each weakness in permissions, enclaves, MCP servers and gateways, verified and ranked by business impact.
Remediation plan
Specific permission, configuration and architecture changes that shrink the blast radius, in priority order.
Controls mapping
Your controls mapped to your governance frameworks, with the gaps marked.
Executive briefing
A board-ready explanation of your AI exposure and the decisions it requires.
How it runs
Scope
A 30-minute call to agree on the AI systems and environments in scope, and a fixed price.
Analyze
Analyze permissions and reach, verify enclave configurations, and analyze MCP servers, API gateways and data controls.
Report
We write the blast radius report: what each AI system can reach, how far a compromise would spread, and recommendations to improve the security.
Brief
The report and remediation plan, presented to technical and executive audiences.
When to call us
Before a deployment goes live
Verify the architecture while it is still inexpensive to change.
When agents gain new access
Every new MCP server, tool or data source changes what a compromise can reach.
When the board asks about AI
Give directors a factual answer about exposure instead of a vendor's assurance.
Questions
What is an AI blast radius?
It is how far the damage spreads if one part of your AI deployment is compromised: the data, systems and workloads that part can reach or change.
Who is this for?
Any organization that runs AI systems, from a single agent to deployments of thousands. It is a great fit for SaaS companies.
Do you need access to our models?
No. This is an assessment of the security architecture around your AI systems: permissions, enclaves, MCP servers, API gateways and data controls.
How long does it take?
It depends on scope. Assessing one AI agent is a different project from assessing 10,000. Your fixed-scope proposal includes the timeline for your deployment.
Related
Services
Advanced Penetration Testing
Adversary-grade testing of networks, applications, cloud and proprietary protocols, combining AI, automation and manual testing.
Training and Briefings
Threat briefings for boards and executives, and hands-on training for technical teams, built around your industry.
Know how far a compromise would spread.
Tell us about your AI deployment. You will get a fixed-scope proposal after a short scoping call.