Why Isn’t AI Transforming Critical Infrastructure (Yet)?

This session will be structured as a facilitated Birds-of-a-Feather (BoF) discussion focused on identifying and characterizing barriers to AI adoption within critical infrastructure sectors. While AI capabilities have matured rapidly, adoption in operational environments (particularly those involving safety-critical OT systems) remains uneven. The purpose of this session is not to showcase a product or a single research effort, but to create a structured forum where practitioners, researchers, and policymakers can surface real-world constraints, unmet needs, and priority use cases.

The session will begin with a brief (10–20 minute) framing overview introducing:

– The NIST AI Accelerator for Critical Infrastructure and its objectives
– The program’s emphasis on trustworthy, operationally aligned AI
– The goal of enabling “machine-speed detection, protection, and response” in cyber-physical environments
– Early pilot focus areas (e.g., AI-enhanced anomaly detection, protocol manipulation awareness, resilience validation in testbeds)

This introduction will establish shared terminology and clarify scope, but will intentionally avoid deep technical detail or marketing content.

The majority of the session (60-70 minutes) will be interactive and discussion-driven. We plan to guide the conversation across several structured themes:

1. Technical Barriers

– Data readiness and quality (e.g., lack of labeled OT data, limited historical telemetry)
– Integration challenges with legacy systems
– Real-time and safety constraints in operational environments
– Model validation, benchmarking, and testbed availability
– Explainability and operator trust

2. Organizational and Governance Barriers

– Risk tolerance and safety culture
– Liability and regulatory considerations
– Workforce readiness and AI literacy
– Procurement and acquisition constraints
– Alignment with standards and compliance frameworks

3. Deployment and Transition Challenges

– Moving from pilot to production
– Demonstrating measurable value (e.g., reduced detection time, resilience gains)
– Vendor interoperability and avoidance of lock-in
– Sustainment and lifecycle considerations
– Participants will be encouraged to share:
– Lessons learned from attempted AI deployments
– Successful early use cases
– Data-sharing constraints and enablers
– Gaps where research or standards guidance would meaningfully accelerate adoption

We will use structured prompts (and, if appropriate, live polling) to ensure balanced participation and to capture themes in a systematic way. The intent is to identify recurring friction points and cross-sector patterns rather than anecdotal commentary.

Expected Outcomes

For attendees:

– Clearer articulation of shared adoption challenges across sectors
– Exposure to emerging use cases and peer lessons learned
– Awareness of opportunities to engage in collaborative pilots or research
– For the broader program:
– Practitioner-informed insight to shape future pilot design
– Identification of priority technical gaps (e.g., data, validation frameworks, testbeds)
– Refined understanding of standards and guidance needs
– Potential collaborators for sector-specific pilots

Target Audience

This session is designed for:

– Critical infrastructure operators (energy, water, transportation, data centers, manufacturing, etc.)
– OT and cybersecurity engineers
– AI researchers working in applied or safety-critical domains
– Standards and policy professionals
– Technology vendors seeking operational alignment

It is not intended to be a technical deep-dive tutorial or a vendor demonstration. Instead, it is a practitioner-informed discussion forum designed to surface implementation realities and shape collaborative next steps.