AI tools are rapidly becoming part of day-to-day workflows across the electric power industry. Engineers are using them to summarize standards, review documentation, analyze datasets, and even assist with compliance checks (the final three of these activities introduces real risk that we'll discuss in this article.)
These tools can provide real productivity gains. But as adoption increases, compliance leaders are beginning to ask an important question:
What happens when sensitive grid and protection system data is entered into general-purpose AI tools?
For organizations subject to NERC requirements — particularly CIP — this is not just a technology decision. It can become a compliance risk.
What Counts as Sensitive Compliance Data?
Many NERC compliance workflows involve information such as:
- Protection system settings
- Facility ratings and limits
- Relay coordination data
- EMS / model parameters
- Transmission topology
- Event analysis results
- Planning and operating studies
- Substation configuration details
Even when this information is not explicitly classified as BES Cyber System Information (BCSI), it may still fall under operationally sensitive information that organizations carefully control.
Uploading this data into public AI tools can create uncertainty around:
- Where data is stored
- Who has access
- Whether data is retained
- Whether data is used for training
- Whether copies exist outside the utility environment
These questions matter — especially when viewed through the lens of CIP requirements.
CIP Considerations When Using Public AI Tools
While applicability depends on each organization's architecture and policies, several CIP areas may come into play when operational data is shared outside controlled environments.
CIP-002 — BES Cyber System Categorization & Identification
Organizations must identify and classify BES Cyber Systems and associated information. If sensitive operational data tied to BES Cyber Systems is uploaded to external AI services, organizations must consider whether that data is being handled consistently with their categorization and protection requirements.
CIP-003 — Security Management Controls
CIP-003 requires documented security policies governing how BES Cyber System information is handled. Use of external AI tools may introduce:
- Unapproved data handling workflows
- Uncontrolled third-party data exposure
- Gaps between policy and practice
If personnel begin using AI tools informally, it can create shadow workflows outside defined security controls.
CIP-004 — Personnel & Training
CIP-004 requires personnel with access to BES Cyber Systems or information to follow defined security practices. As AI tools become more common, organizations may need to clarify:
- What data can be shared
- What tools are approved
- What safeguards must be used
- What constitutes sensitive operational information
Without guidance, well-intentioned employees may unknowingly introduce risk.
CIP-011 — Information Protection
CIP-011 focuses specifically on protecting BES Cyber System Information (BCSI). Depending on the nature of the data, uploading protection settings, topology, or system configuration data into external AI platforms could raise questions around:
- Storage location
- Data retention
- Data disposal
- Access control
- Encryption and transmission
Even when tools claim strong protections, utilities still need to ensure alignment with internal controls.
CIP-013 — Supply Chain Risk Management
CIP-013 requires organizations to evaluate risks associated with vendor products and services. Use of AI platforms introduces:
- Third-party service providers
- Cloud data handling
- External processing environments
- Vendor security posture considerations
This doesn't mean AI cannot be used — but it does mean organizations should evaluate it intentionally.
The Emerging Pattern
We're seeing many utilities take a cautious approach:
- Limiting operational data in public AI tools
- Establishing internal policies
- Using sandboxed environments
- Moving toward controlled, purpose-built systems
The goal isn't to slow AI adoption — it's to adopt AI responsibly.
Where Purpose-Built Systems Help
When compliance workflows run inside purpose-built, closed-loop systems:
- Data stays within controlled environments
- Validation logic is deterministic
- Inputs and outputs are traceable
- Evidence is generated automatically
- Security controls align with utility policies
This allows organizations to benefit from automation without introducing uncertainty around data handling.
As utilities continue exploring AI, the most effective approach is emerging: Use AI to assist knowledge work. data phleet's controlled systems execute compliance validation.
That balance helps reduce risk — while still moving forward.