Development Update: Progress in Post-Deployment AI Observability
Development Update: Progress in Post-Deployment AI Observability
Since 1 April 2026, FASO has made substantial progress in developing its approach to post-deployment AI observability.
The focus of this work remains behavioural and operational change after deployment: how meaningful shifts in the externally visible condition of an AI system may be identified, distinguished and examined over time.
These changes may arise through model updates, changes to surrounding software, policy adjustments, altered orchestration, deployment conditions or other modifications that affect how a system behaves in practice. They may be obvious, but they may also emerge gradually through smaller changes that are difficult to interpret individually.
FASO’s purpose is not to speculate about such changes or to treat every variation as evidence of harm. Its purpose is to improve the disciplined observation of potentially meaningful change while preserving clear boundaries between what has been observed, what has been checked and what may require further assessment.
During the second quarter of 2026, FASO progressed from detailed system design into an implemented Version 1 technical foundation.
This work included the development of governed processes for handling observable indications of change, preserving source and provenance information, distinguishing between different forms of AI system, and maintaining clear separation between observation, analysis, representation and evidential recording.
A significant part of the quarter was devoted to ensuring that the framework does not apply a single undifferentiated method to every form of AI system. Different system types may present different forms of post-deployment change, different observable indicators and different evidential challenges. FASO’s Version 1 work therefore reflects the need for model- and system-sensitive treatment rather than a universal assessment template.
By the end of June 2026, the principal Version 1 implementation had been completed and subjected to extensive automated internal testing. More than 24,000 automated tests were passing with no recorded failures at the completion milestone.
This is an internal engineering achievement rather than a claim of independent validation.
The purpose of the testing programme is to establish whether the implemented system behaves consistently with its own governing requirements, preserves required boundaries and produces reproducible results under controlled conditions. It does not establish external scientific, academic, institutional or regulatory endorsement.
FASO has now moved into a further testing phase focused on the complete operational journey through the system. This work is intended to examine whether the required distinctions and evidential safeguards remain intact from initial observation through to final record formation.
In parallel, FASO has begun preparing for broader external review and institutional development.
The next stage will include discussion of methodological review, governance, long-term stewardship, funding integrity, publication controls and the transition from a founder-led initiative towards a durable public-interest structure.
FASO continues to operate on a non-prescriptive basis. It is not intended to regulate, certify or direct AI developers or deployers. Its role is to support serious, reproducible and verification-conscious observation of post-deployment AI change.
The completion of Version 1 is therefore best understood as the end of the first implementation phase and the beginning of a more demanding period of testing, external challenge and institutional development.