Governing AI inference
AI agents don't escape security controls. They need all five layers.
A defence-in-depth architecture for EMEA enterprises under the EU AI Act, DORA and NIS2.
Kevin Keller ยท kevinkeller.org
Why it matters
Boards, regulators and customers ask the same question
How are you governing the AI systems you're deploying?
The EU AI Act, DORA and NIS2 together make AI governance existential, not optional, for EMEA enterprises.
The stakes
7%
of global annual turnover, or up to EUR 35 million
The EU AI Act's maximum penalty. High-risk AI system obligations apply from 2 August 2026.
The convergence
Three regulations, the same core demands
| Requirement | EU AI Act | DORA | NIS2 |
|---|---|---|---|
| Risk management | Art. 9 | Art. 6-16 | Art. 21 |
| Logging and audit | Art. 12 | Art. 12 | Art. 21(2)(g) |
| Incident reporting | Art. 62 | Art. 17-23 | Art. 23 |
| Access control | Art. 15(4) | Art. 9(4)(c) | Art. 21(2)(i) |
The core principle
You cannot rely on prompts to manage agent behaviour
Data that must not be accessed is blocked by authentication and authorisation, not by instructions to the agent.
Layers 1 and 2
Contain the network, verify every identity
- Network: agents can't call out of, or be called into, their zone
- Region: cross-region inference is disabled by default and can be allowlisted
- Autonomous agents: short-lived service credentials or Workload Identity Federation
- Human-delegated agents: pass the human's identity through, so their RBAC applies
Layers 3, 4 and 5
Authorise outside the agent, stop exfiltration, log everything
- Authorisation: purpose-specific roles, tag-based masking, row access policies
- Data protection: TLS 1.2+, AES-256, no egress except approved endpoints, output scanning
- Audit: immutable logs of every query, tool call and login
You cannot secure what you cannot see.
MCP threats
The tool layer is a real attack surface
| Threat | Impact |
|---|---|
| Tool poisoning | Agent performs unintended actions |
| Confused deputy | Server runs with its own privileges |
| Rug pull | Approved tool turns malicious |
| Exfiltration | Data hidden in legitimate-looking calls |
Pin tool definitions by hash and treat every tool response as untrusted.
Integration choices
From convenience to control
| MCP server | Skill + CLI | Custom agent | |
|---|---|---|---|
| Secrets | Held by the server | CLI's own short-lived auth | In process memory |
| Injection surface | Largest | Reduced | Minimal |
| Audit | Two systems | Two logs | One trail |
| Best for | Prototyping | Production with generic agents | Regulated production |
Defence in depth
Every layer covers another's failure
| Layer | If bypassed by... | Mitigation |
|---|---|---|
| Network | Compromised agent | Egress blocked |
| Authentication | Stolen credential | Short-lived tokens |
| Authorisation | Prompt injection | Role has no grant |
| Data protection | Data in tool parameters | Output scanning |
| Audit | All of the above | Investigation trail |
The takeaway
Enforce every control architecturally: no single layer is enough.
Platform-agnostic principles; practical examples draw on Snowflake's security model