Create
Draft or explore within the authorised framework, without requiring exhaustive fact-checking of every sentence.
Trust infrastructure for artificial intelligence
The full power of AI.
With control built in.
We are developing APLOMB, an independent layer of governance, security and evidence around AI models. When a use case is regulated, sensitive or requires justification, it draws on relevant sources and rules to assess what can be supported.
Power without blind trust
Checks that fit the context.
Not arbitrary restrictions.
Your AI modelsAnthropic · OpenAI · Mistral · the State's Albert gateway · a model installed on your premises
APLOMB · trust layerGovernanceSecurityReliabilityVerifiable evidenceWhat comes out signed certificate, Proof Graph, control log
Your organisationits rules, its data, its final decision
APLOMB does not replace AI models. It coordinates authorised models, tools and sources without unnecessarily reducing their capabilities.
Access rights and confidentiality still apply. Verification becomes more extensive when requested, required by the stakes or mandated by the organisation’s rules.
Verify
Examine the requested claims, retrieve their sources and identify what remains uncertain.
Regulated use
Apply the checks and approvals defined by the relevant policies before the use or distribution concerned.
These are not three security levels: no profile disables access rights, confidentiality or mandatory organisational rules.
One ecosystem, distinct roles
A shared foundation.
More than one way to use it.
The APLOMB foundation
Infrastructure being built around AI, tools and sources: coordinating use, applying rules and retaining evidence.
Understand the foundationWorkspace and APIs
An interface for users and integration paths for third-party software. Two ways to use the same foundation, depending on the capabilities actually connected.
Explore integrationsAPLOMB LAB
Applied research and innovation: inventing, prototyping, challenging and preparing APLOMB’s next capabilities.
Explore the researchTarget architecture. Availability must be confirmed for each capability, version and scope.
Governance · Security · Reliability · Verifiable evidence
Four requirements.
One framework.
Coordinating authorised AI models and tools, bringing useful sources together and applying appropriate checks: the intended role of APLOMB’s orchestration and aggregation.
Governance
Who may use which AI?
Choose authorised AI models, define uses and budgets, and specify who decides and when human approval is required.
Understand the infrastructureTeamsSoftwareAI agents
/.PLOMBThe shared framework
AI modelsToolsSources
Your organisation’s rules · Conceptual view
Security
Which data and which actions?
Control access, data exchanges and the tools involved. An incoming instruction is not permission: operations remain subject to the organisation’s rules.
Understand the infrastructureTeamsSoftwareAI agents
/.PLOMBThe shared framework
AI modelsToolsSources
Controlled access and operations · Conceptual view
Reliability
What supports the result?
Use the reference corpus, examine applicable passages and rules, seek contradictions and make uncertainty visible. Checks depend on the context and the sources actually available.
Understand the infrastructureTeamsSoftwareAI agents
/.PLOMBThe shared framework
AI modelsToolsSources
Results to examine, limits to understand · Conceptual view
Evidence
What can be checked afterwards?
Link a check to its sources, the content version and its limits. Verifiable evidence means being able to examine what was checked, without equating a signature with overall accuracy.
Understand the infrastructureTeamsSoftwareAI agents
/.PLOMBThe shared framework
AI modelsToolsSources
Identified sources, versions and checks · Conceptual view
The APLOMB reference corpus
Evidence begins
with the right references.
We are enriching a reference corpus of official texts, domain references and structured rules. It is intended to test claims against material relevant to the jurisdiction, date and use case.
Official texts · Domain references · Organisational rules
Understand evidence and its limitsQualify the source
Identify its origin, authority, version and scope.
Examine what it says
Locate the relevant passage, its conditions, exceptions and any conflicting evidence.
Support or qualify the conclusion
Retain the evidence from the checks. Where evidence is missing, the conclusion must remain qualified or unestablished.
Corpus coverage and freshness must be documented by domain. Finding a reference does not establish an entire answer; a signature does not guarantee its accuracy.
An example in a local authority
Your work.
Not just an answer.
One interface to the APLOMB foundation: create a document, retain its references and revisit changed passages. This example illustrates the link between creation and evidence; it does not define the whole infrastructure.
Explore the experience ↗APLOMB WorkspaceLOCAL DEMONSTRATION
Note · digital servicesv1
DocumentExample with a reference
Preparing accessible
digital services.
Working note · demonstration data
The demonstration inventory lists 80 workstations.
The proposed priority is to support teams and identify user needs before deciding on deployment.
Passage edited. The reference remains linked. The previous version’s check does not automatically apply.
[Service manager to be completed]
01Demonstration inventory
Fictional source: “The inventory contains 80 workstations.” This scenario demonstrates viewing a reference; it does not perform a real check.
Illustrative version 1 · no data saved
LINKS IN THE TEXTTARGETED EDITSSEPARATE VERSIONS
Prewritten interactive scenario. No model calls, no storage and no certificate issued.
01Edit in context
The document and conversation stay within reach.
02Keep references
The source stays linked to the passage it informs.
03Distinguish versions
An edit does not inherit an old validation.
APLOMB LAB
Designing and testing
trust.
APLOMB LAB is our applied research and innovation effort. We design and test new ways to govern, secure and verify the use of AI. Ideas become prototypes, challenged against errors, contradictions and practical use cases; their results and limitations guide which capabilities should be integrated into APLOMB, with an emphasis on frugal engineering and reproducibility.
Discover the LAB ↗APPLIED RESEARCH · APLOMBPractical questions
Ideas open the way.
Tests decide.
The foundation governs use. The LAB develops and tests its next capabilities.
APLOMB LAB researches and prototypes new control, verification and evidence capabilities. It examines source qualification, domain rules and resource efficiency, then prepares the integration of ideas that withstand testing.
01Source relevanceDoes a real reference support the claim?
02Evidence continuityWhat needs reviewing when a document changes?
03Efficient verificationHow can we check better without repeating everything?
Proposed research areas, not announced scientific results.
Our first application areas
The need for evidence
extends beyond one sector.
Local authorities, public administrations and government departments are our initial priority. APLOMB also targets organisations, regulated sectors and software whose AI use requires checks and justification.
PUBLIC SECTORLocal authorities &
Organisations &
Partners &
Local authorities &
administrations
Define sensitive AI uses, data access, applicable rules and human approval.
Scope a pilot ↗INTEGRATIONOrganisations &
software vendors
Explore integrating the same trust framework into business and regulated environments, subject to available capabilities.
Explore integrations ↗RESEARCHPartners &
experimenters
Challenge methods and develop tests around practical use cases.
Meet the LAB ↗Is APLOMB a new AI model?
No. APLOMB develops independent orchestration, governance and evidence infrastructure. The Workspace is one way to use it, not its entire definition.
Is APLOMB only for the public sector?
No. Local authorities, public administrations and government departments are our initial priority, not a functional boundary. APLOMB is developing infrastructure that other organisations and software can reuse, subject to the integrations and capabilities actually available.
Does APLOMB need to verify everything, all the time?
No. Access rights, confidentiality and mandatory safeguards still apply. The depth of verification depends on the need, risk and applicable policy. A user may request additional verification, but cannot remove an obligation.
Does an enriched corpus guarantee a correct answer?
No. The relevant references must be selected and checked against the claim in its context. Conflicting, outdated or insufficient sources must lead to qualifications or an unestablished conclusion, not a false verified status.
Does governance restrict creativity?
The goal is to preserve supported and authorized capabilities. Privacy rules, budgets and permissions still apply; limitations must be explicit, never silent.
Is a signed document necessarily accurate?
No. A signature helps verify integrity within a defined scope. Claims, sources and the checks actually performed must be assessed separately.
Can all the presented workflows already be tested?
The illustrations show the intended experience. Availability is confirmed for a tested version, connection and environment. The pilot’s scope is defined before real use.
Our signature · AI generates. APLOMB establishes what can be proven.
Adoption · Integration · Research