ENNL
Book a call
All products

Products / Agent security hardening

Agents that stay inside their limits.

We implement the defence layers around your AI agents: an identity per agent with least-privilege tools, allow lists, approval gates for actions with impact, sandboxed execution, limits per agent and per swarm, a kill switch, tracing with alerts and an attack evaluation set that gates every release.Scope: Scope agreed per system

Book a callAI security engineering

01 / Who it is for

Who it is for

  • ForTeams with agents in production or close to it, which need their risks reduced and their controls written down.
  • The problem it solvesAgents hold broad rights, act on text they read and run without a hard limit on spend or steps, so one bad instruction can become one bad action.

02 / What you get

What you get

The deliverables of this product. Nothing else is in scope until we agree it.

  • Threat model of your agents against the OWASP LLM Top 10 and the agent-specific threats
  • Identity per agent, least-privilege tools and allow lists
  • Approval gates, limits, kill switch and circuit breakers
  • Tracing with anomaly alerts
  • An attack evaluation set that runs as a release gate
  • Runbooks for incidents

03 / How it runs

How it runs

Fixed steps in a fixed order. Scope: Scope agreed per system.

  1. Model the threats

    We map each agent, its tools, its data and its trust relations, and rank the threats.

  2. Design the layers

    We choose the controls per threat and write the design, with an owner for each control.

  3. Implement

    We build the controls as code in your repositories and deploy them through your pipelines.

  4. Test and gate

    We build the attack evaluation set, wire it into the release pipeline and run it against the hardened agents.

  5. Hand over

    We write the runbooks and train your team to run the controls and extend the attack set.

04 / Your result

What you hold at the end

Agents with the defence layers in place, an attack evaluation set that gates releases, and runbooks your team can run.

What we need from you

  • Access to the agent code, its tools and its configuration
  • A named owner for each agent
  • A release pipeline we can add a gate to
  • One engineer who will own the controls afterwards

Scope fixed before we start; quote after a first call.

06 / FAQ

Questions about this product

Do we need to rebuild our agents?

No. We add the layers around what you run: identity, tool access, limits, approvals and tracing. Where an agent needs a change, we make it in your code.

Does this work on Databricks?

Yes. We use Unity Catalog for access, Unity AI Gateway for spend and logs, and your pipelines for the release gate. The same layers apply to agents on other platforms.

Can a penetration test come first?

Yes. A test shows which layers matter most, and the hardening then closes the findings. We run them as two separate scopes.

Next step

Start with Agent security hardening.

Book a call and we agree the scope in writing. We reply within two working days and say plainly whether this product fits.

Book a callSend a messageor email info@nexops.nl directly

Currently accepting projects for Q4 2026