Where AI governance meets operational reality.

PHYSICAL AI

When AI can act, evidence must follow the action.

Physical consequences extend the question beyond a model output. Examine who authorised an action, what constrained execution, which system state was active and how intervention or recovery was evidenced.

The scope follows a defined physical action path. Security assurance and functional safety remain distinct.

Explore the evidence questions →

ACTION BOUNDARY

Examine the transition from decision to effect.

Follow the point where an AI decision can change an actuator, environment or other physical state. Authorisation, policy gates, vetoes and override paths belong in the evidence record.

An accepted command is not, by itself, evidence of the physical result. Where the question concerns motion or intervention, distinguish the requested action, execution acknowledgement and observed physical behaviour.

EXAMINATION QUESTIONS

Four records at the point of consequence.

The required record depends on the specific system, action pathway and question being examined.

01 / AUTHORITY

What authorised the action?

Identify the actor or system, delegated permissions, applicable policy and approval context. Preserve the basis for authority at the time of the action.

02 / BOUNDARY

What constrained execution?

Identify gates, limits, denials and vetoes. Distinguish a configured restriction from evidence that it operated in the relevant conditions.

03 / STATE

Which system produced it?

Link the record to the version, configuration, environment and operating state. Keep timestamps and observation limits explicit.

04 / RECOVERY

What shows intervention?

Examine records of containment, override, physical response and return-to-service decisions where relevant. State what was observed and what remains unverified.

FRAMEWORK CONTEXT

PAI-SF™ provides the security architecture.

The Physical AI Security Framework applies Kinetic Zero Trust principles across sensing, decision-making, authorisation and actuation. Consult the published control text for its version, applicability and scope.

Framework context helps define what to examine. A mapping or documented control does not establish that a physical system behaved as intended.

DISTINCT INTERFACES

Security assurance does not replace functional safety.

Sector-specific approval, functional-safety obligations and operational authorisation remain separate. The scope and competence required for each must be established explicitly.

Notably Absent

  • No release approval, safety certification or sector authorisation implied.
  • No simulation result treated as proof of real-world protective behaviour without an established basis.
  • No general safety or compliance conclusion beyond the examined record and scope.
  • No claim of proprietary telemetry, completed physical tests or client outcomes made here.

DISCUSS A SYSTEM

Bring one physical action path and its records.

A useful first note identifies the consequential action, authority boundary, system context and types of telemetry or decision records available.

Share a non-confidential description initially. Externally directed analysis and examination require an agreed paid scope, relevant competence, independence checks and authorised data access. Do not send credentials or sensitive operational files in the first enquiry.

Discuss a Physical AI question →

SERVICE ROUTES

Bring the physical action path into scope.

A Physical AI question can enter through Applied Research or Assessment Services, depending on the decision, evidence and examination required. These are existing service routes, not a separate certification offer.