Guardrails — Status Quo vs What's Needed
Last edition, I had covered why evals are not audits. Claude rephrased it as — evals are vibe audits. And it's only natural to ask — can we do an audit at runtime and catch policy violations? That's what guardrails are supposed to be.
It's always important to ask — how were things done before Nov '22? Did we even have guardrails before GPT?
Guardrails Are Not New
Of course we did. We just didn't call them that. Remember — how sign-up forms required a business email and Gmail IDs won't be accepted? That was a classic marketing guardrail.
- Every claims processing system had validation rules.
- Every EHR had required fields and range checks.
- Every prior auth workflow had decision trees that enforced policy before a human saw the output.
If a dosage fell outside the approved range, the system rejected it. Not probabilistically. Deterministically.
These weren't sexy. Nobody built a pitch deck around them. But they worked. A business rule that says "pediatric dosage cannot exceed X mg/kg" doesn't hallucinate. It doesn't drift.
What Actually Changed in the Post-GPT World
"Both input and output went from structured to unstructured. The guardrails that worked on structured data have nothing to grab onto."
When faced with a hard question, we answer an easier one instead, and don't notice we've done it.
Daniel Kahneman
The Substitution Effect
Fig 2"Does this AI output comply with our SOPs?"
Requires: reasoning over policy logic, version checks, patient-category matching. Slow, deliberate, System 2 work.
Needs: formal knowledge structure, ontology, versioned policy rules.
Guardrails Are THE Feedback Layer
Here's the thing most teams miss: guardrails aren't just a safety or compliance enforcement system. They provide feedback and set the foundation for a closed-loop learning system.
The point isn't just to block bad outputs. It's to catch what went wrong, figure out why, and feed that back into the system so it gets better.
- Shallow feedback: "Toxic content detected" → block the response, move on. System learns nothing useful.
- Rich feedback: "Response cited Protocol v2021 when v2024 changed the threshold from 5.0 to 7.5. Knowledge base needs update." → System learns exactly what to fix.
The first triggers a whack-a-mole feeling. The second is an actual learning loop. Most "guardrails" are in the first layer. Regulated industries need the second.
Most guardrails are shallow. Regulated industries need the rich feedback loop.
Guardrails = Real-Time Audits
If evals are post-deployment quality checks, and audits are post-deployment compliance proof — then guardrails should be audits done in real time, with enforcement. Not "did the AI say something harmful?" but "did the AI follow Protocol X, Section 4.2, using the correct version?"
The Guardrails 2×2
| Timing | Basic Quality | SOP Compliance | Coverage Gap |
|---|---|---|---|
| Runtime (before response) | |||
| Post-deployment (after the fact) |
Meet Human-In-The-Loop
"Don't worry, we have a human-expert-in-the-loop." This is the phrase that ends the safety conversation too early. And to be fair — HITL makes sense as a bridge.
But here's where it breaks: human-in-the-loop distributes liability without solving the underlying problem. Throughput beats accuracy. Every time.
We expanded this into a full diagnostic: Phantom Human-In-The-Loop →
The Alternative: Neuro-Symbolic Guardrails
Guardrails need to be grounded in a source of truth — not 1,000 embeddings of chunks of text.
The LLM handles what it's good at — interpreting unstructured language, extracting clinical intent. Then a symbolic reasoning layer — ontologies, decision graphs, versioned policy rules — checks that intent against your actual SOPs. Deterministically. With traceability.
That's how you get from "does this sound right?" to "does this provably follow the right protocol?" — which is what guardrails were always supposed to do.
Explore the architecture: Neuro-Symbolic AI — A Practitioner's Taxonomy →
Vivek Khandelwal
2X founder who has built multiple companies in the last 15 years. He bootstrapped iZooto to multi-millons in revenue. He graduated from IIT Bombay and has deep experience across product marketing, and GTM strategy. Mentors early-stage startups at Upekkha, and SaaSBoomi's SGx program. At CogniSwitch, he leads all things Marketing, Business Development and partnerships.