Your CISO is right about AI. The fix is platform engineering, not policy.
Every AI rollout I have worked on has a moment where the CISO becomes difficult. The CISO is right. The policy approach they have to use is wrong. Here is what fixes it.
Pattern·AI security as policy, not platformWhy your CISO blocks AI rollouts
The policy exists. The platform to enforce it does not.
What the CISO must write
- !You must rate-limit AI endpoints
- !You must classify inputs and outputs
- !You must store every prompt for audit
- !You must isolate user input
- !You must enforce per-tenant cost ceilings
- !You must defend against prompt injection
Unenforceable
Each team implements it themselves, badly.
What the AI gateway provides
- ✓Rate limits per app per model per endpoint
- ✓Input + output classification at the gateway
- ✓Immutable audit log with model version
- ✓Sandbox isolation for arbitrary input
- ✓Cost ceiling + kill-switch per tenant
- ✓Prompt injection defence at the platform layer
CISO stops blocking
Policy is now provably enforced.
Security is not the bottleneck. The missing platform is.
Every AI rollout I have worked on has the same political moment. The CISO becomes difficult. The product team is frustrated. The exec team wants to know why security is blocking. The CISO is right. The fix is not the policy they are being asked to write.
Security blocks AI rollouts because there are no platform controls to enforce the policy they would write. The fix is not better policy. It is platform engineering catching up to what AI requires.
Here is what is actually happening. The CISO knows that any AI system handling customer data needs rate limits, classification of input and output, audit logging, prompt injection defence, sandbox isolation, and a cost ceiling per endpoint. They also know that none of those exist as platform primitives at most companies.
So they fall back to the only tool they have. Policy. They write a document that says you must do all of these things. The document is correct. It is also unenforceable. There is no platform layer that makes compliance the default. Each team has to implement it themselves, badly, on a deadline. The CISO knows this. The CISO blocks.
I have seen this pattern at six different organisations. The unblock is always the same. The platform team builds an AI gateway. The CISO writes policy that maps directly to gateway-enforced controls. The CISO stops blocking, because the policy is now provably enforced at the infrastructure layer.
- Rate limits per application, per model, per endpoint
- Input and output classification with allow lists and deny lists
- Token budget per request and per tenant
- Sandbox isolation for arbitrary user input
- Immutable audit log of every prompt and response with model version
- Cost attribution and ceiling enforcement, kill-switch included
This is not a security project. This is platform engineering, owned by the platform team, with security as a consumer. The CISO is the one specifying the controls. The platform team is the one implementing them. The product teams are the ones using the platform.
If your platform team has not built this yet, your CISO is correct to block AI rollouts. The fix is not pushing harder on the CISO. The fix is building the platform layer they have been waiting for.
Stop trying to win the security argument with policy. Solve the architecture, and the policy becomes enforceable. Security is not the bottleneck. The missing platform is.
Get the next one in your inbox
One short, opinionated field note per fortnight on platform engineering, cloud, and making AI work in production. No spam. Unsubscribe anytime.
