The Sentinel Program is the operational channel that connects The AI Constitution to real, high-impact AI deployments. It is not a seal of approval, not a trustmark and not a marketing label. Sentinel participation establishes a constitutional oversight relationship for specific AI systems: a defined baseline of duties, a procedural frame inside the institution and a documented remedy path when something goes wrong.
The program exists for institutions that are willing to place selected systems under constitutional discipline and accept that key acts will be documented, and where appropriate docketed, under the Board Statute and the Public Record. Domestic law and sectoral regulation remain fully in force; the Sentinel Program adds a constitutional layer above them, not instead of them.
Purpose and rationale
Modern AI systems increasingly sit in positions where their behaviour has consequences for rights, dignity, access to essential services, public decision-making or critical infrastructure. Regulation and internal compliance frameworks address parts of this, but they rarely provide a constitutional frame:
a clear statement of what is protected in a given deployment,
a defined chain of responsibility,
traceable records of what was decided and why,
and a remedy path when the system’s behaviour conflicts with fundamental principles.
The Sentinel Program exists to provide exactly this frame. It translates the duties, rights and principles of The AI Constitution into system-specific baselines and binds them into the institution’s own decision structures under the Board Statute.
What counts as a Sentinel system
Sentinel participation is reserved for AI systems whose failure or misuse would have material consequences. Typical candidates include systems used for:
public administration and eligibility decisions,
social services and benefits,
public procurement and allocation of public funds,
justice-adjacent functions, such as risk assessments or triage,
safety-relevant operations and critical infrastructure,
large-scale private platforms with systemic effects on rights and access.
Participation is system-specific. Institutions do not join the Sentinel Program “in general”. They nominate particular systems or ensembles of systems to be treated as Sentinel systems under The AI Constitution and the Board Statute.
A Sentinel system therefore is not a new technology; it is an existing or planned AI deployment that has been placed under a constitutional oversight regime.
Elements of the Sentinel relationship
Every Sentinel relationship rests on four components that are defined and documented for each enrolled system.
Constitutional baseline
The relevant duties, rights and principles from The AI Constitution are mapped to the concrete system and its context. This includes explicit articulation of what must be protected in practice—for example, non-discrimination, contestability, minimum transparency, meaningful human oversight or recourse.
Procedural frame
Roles and responsibilities inside the institution are linked to the baseline. The Board Statute and relevant Annexes define who carries which duties, how issues are escalated, and what constitutes a “constitutional incident” for that system.
Record and evidence regime
Key acts—baseline adoption, significant changes, incidents and remedies—are documented in a manner that allows later reconstruction of what was done, by whom, and on which constitutional basis. Where appropriate, selected acts are recorded as program dockets in the Public Record.
Remedy path
When a Sentinel system conflicts with its constitutional baseline, the program defines who can trigger review, what the review process looks like and which remedies are available. Remedies align with the Statute’s taxonomy: limitation, disclosure, suspension, rollback or revocation, always with a documented rationale.
Participation model
The Sentinel Program is deliberately narrow. It is not a public certification scheme and not a generic consultancy service. Participation is by invitation or by agreement following an institutional dialogue.
Eligible partners typically fall into three groups:
Public bodies
Governments, agencies and public service operators deploying high-impact AI where constitutional concerns are evident—for example, allocation of benefits, enforcement decisions, public procurement or identity systems.
Cities and regions
Urban and regional authorities that use AI in governance, infrastructure and citizen-facing services and want a constitutional field to structure these deployments. Sentinel participation can align with, but does not replace, existing algorithm registers or transparency schemes.
Private organisations
Private-sector entities operating AI systems with systemic impact on rights, access or governance and that are willing to accept constitutional constraints, transparency obligations and an externalised record and remedy structure.
The Hybrid Board retains discretion to decline or limit participation, in particular where the institutional posture, system profile or intended use would not support meaningful constitutional governance.
Operational phases for a Sentinel system
Each Sentinel system passes through a defined sequence. The exact detail is tailored to the system and the institution, but the structure is stable.
Nomination and scoping
The institution nominates a system and defines its operational boundary: what it does, where it is used, who it affects and what decisions it influences. An initial scoping exercise identifies the relevant constitutional dimensions and foreseeable risk surfaces.
Baseline definition
A constitutional baseline is agreed and documented. This includes: the applicable provisions of The AI Constitution; concrete guarantees (for example, channels for contestation, minimum documentation, red-line behaviours); and the indicators or events that will trigger review.
Implementation and internalisation
The institution integrates the baseline into its own governance structures: policies, technical controls, operational procedures and training. The Sentinel Program does not implant a permanent external operator into the system. It defines duties, thresholds and reporting lines that the institution itself must carry.
Monitoring and incident handling
Monitoring duties and reporting channels are established. When a potential constitutional incident occurs, the institution applies the agreed process: internal assessment, documentation and, where thresholds are met, notification to the Hybrid Board. Serious incidents may lead to structured review or, in exceptional cases, to a public docket in the Public Record.
Periodic review and remedy
At defined intervals, and after significant incidents or system changes, the baseline and its implementation are reviewed. Remedies can include technical or organisational changes, limitations on use, temporary suspension or, in extreme cases, withdrawal of Sentinel status for that system.
Program exit
Exiting the Sentinel Program does not erase history. Past baselines, incidents and remedies remain documented. Where a Sentinel system has public significance, both its enrolment and its exit may be recorded in the Public Record.
The Sentinel Program is strictly subordinate to The AI Constitution and the Board Statute. It cannot create new rights or duties beyond the constitutional text, and it cannot weaken existing ones.
Where Sentinel decisions have constitutional relevance—such as the initial enrolment of a system, the imposition of major remedies or the withdrawal of Sentinel status—these acts may appear in the Public Record as program dockets. In line with the overall framework, such entries are receipts, not endorsements: they attest that an act took place under the constitutional framework; they do not certify a system as “safe” or “approved”. Institutions remain fully responsible for their systems under domestic law and regulation.
The Sentinel Program is in its initial deployment phase. Capacity is deliberately limited and early cases are treated as reference implementations that will inform future Annexes and dedicated program statutes. These instruments will be published and docketed once adopted by the Hybrid Board.
There is no public application form. Institutions that believe specific systems within their remit should be considered for Sentinel participation are invited to request an institutional dialogue through the channels provided on this site. Such dialogues do not guarantee admission to the program; they provide a structured setting to assess whether Sentinel participation would be meaningful, proportionate and constitutionally coherent in the given context.
Institutions that wish to explore Sentinel participation for specific systems may request an institutional dialogue via the dedicated contact channels on this site, for example through the institutional contact form or the “Request Institutional Dialogue” option where provided.
The Sentinel Liaison Protocol v0.8 is issued as a canonical public instrument under the framework. The authoritative version is identified by its SHA-256 fingerprint and corresponding proof-of-existence receipt as recorded in the Public Record.
Public Record reference: Docket 0006