AI agents and robots change a mining threat model in three ways: more software you did not write sits near physical processes, data flows from sites to cloud models, and a model can now propose or take actions. Existing operational technology guidance still applies, and recent joint guidance adds AI-specific principles.
We have not tied this post to any specific incident, and we do not claim any mine has been attacked through an AI system. This is a planning view built from published guidance.
What does the published guidance say?
NIST SP 800-82 Revision 3, published in September 2023, is the reference guide for operational technology security. It defines OT as programmable systems and devices that monitor or control physical processes and equipment, and it covers typical topologies, common threats and recommended countermeasures while accounting for performance, reliability and safety.
In December 2025 CISA, the Australian Signals Directorate's ACSC and international partners published joint guidance on integrating AI into operational technology. Per the Canadian Cyber Centre's summary, it sets out four principles: build AI literacy, evaluate where AI fits in OT, set up governance and assurance, and design in safety and security. CISA's page says the guidance covers machine learning, large language model based systems and AI agents, and encourages owners to continuously monitor, validate and refine their models.
The ISA/IEC 62443 series splits responsibility among asset owners, product suppliers, integrators and service providers, and part 4-1 covers secure product development lifecycles. In a mine, that shared model is real: equipment makers, integrators and operators each hold part of the answer.
What changes with AI agents and robots?
More software, further from your control. Autonomous haulage, drilling and monitoring systems arrive with firmware, models and open source libraries from several suppliers. You may not know what is inside.
New data paths. Models often need telemetry, which means data leaves the site or a gateway collects it. Every path is an attack surface and a privacy question.
Decisions, not just displays. An agent that recommends a setpoint is different from a dashboard. If a model's output can reach a controller, a wrong or manipulated output has a physical consequence. This is why separating advice from action matters.
Remote access grows. Vendors want remote support and updates. Each remote path needs authentication, logging and an owner.
Safety and security interact. A control that protects against intrusion, such as automatic lockout, can interfere with a safety action. Design reviews need both viewpoints.
What should a threat model include?
Start from the physical process, not the IT diagram. List what could go wrong physically, then trace which software and data paths could cause it. Add the AI components as separate items: model, data source, tool access and the human who approves actions.
Ask for software bills of materials from suppliers, so you can see libraries inside vendor systems. When a library flaw is announced, the SBOM tells you which systems to ask about.
What to do
- Inventory OT assets, including autonomous machines, gateways and remote access paths.
- Request SBOMs and update commitments from each equipment and software supplier.
- Draw data flows from site to any cloud or model service and decide what must stay local.
- Keep model outputs advisory unless a human or a hard safety limit sits between them and a controller.
- Log model inputs, outputs and approvals.
- Include AI components in incident response plans and exercises.
- Read NIST SP 800-82 and the joint AI in OT guidance with your engineering and safety leads together.
Quick check
Does AI replace existing OT security practice?
No. The joint guidance builds on existing OT practice, and NIST SP 800-82 remains the base reference.
What is the first control to add?
An accurate inventory of assets, software and remote access paths. Every other control depends on it.
Sources
- NIST SP 800-82 Revision 3, Guide to Operational Technology Security
- CISA: Principles for the Secure Integration of AI in Operational Technology
- Canadian Centre for Cyber Security: joint guidance summary
- ISA/IEC 62443 series overview
Last reviewed: 18 October 2026.
