FASO Glossary

Accountability
The responsibility to preserve auditable records, controlled processes and documented review.

AI system
A model, service or technical arrangement that produces Automated Intelligence -mediated outputs or actions.

Audit-bound
Designed so that observations, records and decisions remain traceable and reviewable.

Behavioural drift
A meaningful change over time in the externally visible behaviour of an AI system.

Bounded observation
An observation presented within defined limits, including what is known, what remains uncertain and what the evidence does not support.

Confidence warranted
The level of confidence justified by the available evidence.

Custody
The controlled handling and preservation of evidence or records through a process.

Deterministic process
A process designed to produce consistent results when the same authorised inputs and conditions are used.

Discovery
The identification of potentially relevant signs, events or changes for later examination.

Evidence
Information preserved in a form that can be checked, reviewed and traced to its source.

Externally visible condition
A feature or behaviour of an AI system that can be observed without privileged internal access.

FASO
The Frontier Alignment and Safety Observatory, an independent and neutrality-preserving observatory concerned with post-deployment AI change.

Frontier AI
Advanced AI systems near the leading edge of current capability or deployment.

Governance posture
The principles and structural position through which FASO preserves neutrality, independence and bounded operation.

Human review
The examination and interpretation of system-produced records or outputs by trained people.

Independent observatory
An organisation that observes and reports without acting as a regulator, command centre, commercial operator or enforcement body.

Neutrality-preserving
Structured to avoid advocacy, prescription, commercial capture or improper influence over findings.

Non-prescriptive
Not directing developers, deployers, regulators or other actors on what action they must take.

Observability
The ability to identify and examine externally visible signs of system condition or change.

Phase 0
FASO’s controlled build-and-proof stage, focused on defining, testing and reviewing its core method before wider operational scale-up.

Post-deployment
The period after an AI system has been released or placed into operational use.

Provenance
Information showing where evidence or records came from and how they were handled.

Publication Boundary
The controlled boundary governing what FASO may safely and responsibly publish externally.

Publication-safe
Suitable for external release without exposing sensitive operational detail, private information or material that could enable misuse.

Replay discipline
The preservation of sufficient records and process controls to allow earlier handling or analysis to be reconstructed and checked.

Reproducible
Capable of being repeated under the same conditions with a consistent result.

Ring-fenced funding
Funding reserved for a defined purpose and protected from use outside that purpose.

Threshold
A predefined point at which an observation may require additional governed review or escalation.

Verification
The process of checking whether information or a recorded condition is supported by the available evidence.

Visualisation
A bounded presentation of evidence or analytical state designed to show what changed, what is supported and where uncertainty remains.

Wrapper
Software or configuration surrounding an AI model that may affect how it behaves or interacts after deployment.