We Submitted Feedback to NIST on AI and Zero Trust - Here’s What We Said
- Kristopher Persad
- Mar 2
- 3 min read
The future of cybersecurity will be shaped by how we integrate AI, not just how we defend against it.
Recently, NIST released the Initial Public Review Draft of IR 8596: Cybersecurity Framework Profile for Artificial Intelligence. The document aims to extend the Cybersecurity Framework to address both the security of AI systems and the use of AI in cybersecurity operations.
We reviewed the draft closely and submitted formal public comment.
Here’s our take.
What NIST Got Right
The draft does an excellent job acknowledging two critical realities:
AI systems must be secured like any other high-value asset.
AI will increasingly be used to enhance detection, response, and cyber defense capabilities.
It recognizes AI supply chain risk, model integrity, and adversarial manipulation. It also anticipates AI-enabled attacks and automation at scale.
That’s a strong foundation.
But we believe there’s a structural gap that must be addressed as AI becomes embedded in operational security workflows.
The Missing Piece: AI as an Actor
Most frameworks today treat AI as either:
A system to protect, or
A capability to leverage
What they rarely address explicitly is this:
What happens when AI takes action?
Modern AI security capabilities are already:
Correlating telemetry
Proposing configuration changes
Triggering workflows
Disabling accounts
Initiating remediation steps
Acting with limited or no human interaction
At that moment, AI is no longer just a tool. It becomes an actor. And actors require identity.
Why Identity Matters
Zero Trust is built on the premise that every actor must be Authenticated, Authorized, Scoped, Logged, and Revocable.
Historically, those actors were either:
Humans
Service accounts
But AI systems performing operational functions don’t fit neatly into either category - yet functionally, they behave like both.
Our feedback to NIST centred on this position:
AI agents, copilots, and autonomous workflows used in cybersecurity operations should be treated as non-human identities.
Without explicitly modelling AI systems this way, organizations risk:
Over-permissioned AI capabilities
Weak auditability of AI-initiated actions
Blurred accountability
Erosion of Zero Trust enforcement
As AI systems increasingly operate at machine speed, identity governance must evolve alongside them.
Authorization Is No Longer Binary
Another dimension we highlighted is authorization granularity.
AI systems today may:
Read and analyze
Recommend actions
Execute with human approval
Execute autonomously under defined conditions
Those tiers matter.
Treating AI authorization as simply “enabled” or “disabled” is insufficient. Decision-tiered authority will become essential as AI adoption matures.
This Is About Governance, Not Restraint
Our position is not anti-automation. It’s pro-accountability. AI-enabled security operations can dramatically improve speed and resilience. But as soon as AI influences or executes actions inside an enterprise environment, identity and delegation models must be explicit.
Frameworks must evolve from:
“How do we secure AI?”
to
“How do we govern AI as an operational participant?”
Where This Goes Next
NIST’s draft is an important step forward. It signals that AI governance and cybersecurity are converging and that’s necessary.
But as enterprises deploy semi-autonomous and autonomous AI capabilities, we believe frameworks must formally recognize:
AI systems as actors
Actors as identities
Identities as governed entities
Zero Trust reshaped how we think about user access.
AI will reshape how we think about operational authority.
The sooner our frameworks reflect that reality, the more durable they will be.
If you’re navigating AI adoption inside your security organization, this conversation is just beginning. And it’s one we should be having now, not after governance gaps become incident reports.
Comments