Trust
Privacy and civil liberties. Trust is not promised: it is built.
Edelon is founded on a conviction: an institution can use its data to make better decisions without weakening people’s rights.
We build software for decisions with consequences, in defense, in the public sector and in companies that run what is essential. There, a design error does not stay on the screen: it can reach someone’s liberty, security or property. That is why privacy and civil liberties are not a review at the end of the project, but an engineering condition from the first line of code.
From their first day, everyone at Edelon learns to ask one question before building:
What would someone who should not have this do with it?
Principles
As our software answers questions of greater reach, we bind ourselves to rules that do not change from one customer to the next or from one moment to the next.
- Privacy is designed in, not added on.
- We treat privacy as a first-order requirement at every stage of engineering. Protections are core capabilities of the platform, built into the same tools used to analyze and collaborate, so that protection does not cost utility.
- Decisions that affect rights are not left to a machine alone.
- When a piece of data can weigh on a person’s liberty, opportunity or security, the system allows the decision to be reserved for someone with the authority to make it, with the necessary context and with a path to correct it.
- Every system must be accountable.
- We design EDELON OS so that every access, every action and every decision leaves evidence that those who exercise oversight can review and reconstruct, without relying on the word of the operator or the vendor.
- A fact, an inference and a proposal are not the same thing.
- What is confirmed, what the system infers and what an AI suggests travel as distinct categories. An inference does not establish a fact, and a recommendation does not authorize an action.
- Each organization owns its data.
- Isolation between organizations extends to queries, derived results, exports and integrations, not just the screen. Sharing requires explicit authorization, and that authorization can be revoked.
- Technology is not the answer to everything.
- Some decisions are too grave to be automated, even with a person in the loop. When we cannot build a responsible solution, we decline the opportunity. The uses we do not support are set out in our Human Rights Policy.
How these principles are met changes with each product and with each institution. The aim does not change: to build technology with full awareness of its effects on fundamental rights.
Technologies
EDELON OS is designed so that data protection, governance, the responsible use of AI and the protection of civilians are functions of the system rather than manual processes. In this way, each institution can meet its obligations, from the European General Data Protection Regulation and AI Act to international humanitarian law and the law of each jurisdiction where it operates.
- Granular access control and dynamic minimizationBastion
- Permissions by role, attribute, purpose and duration: each person sees only what is necessary for a defined task, and only while performing it. On top of access, pseudonymization, masking and encryption reduce what is exposed according to the context of each flow.
- Detection and management of sensitive dataBastion
- Markings of sensitivity, origin, purpose and expiration accompany the data from its source to objects, flows, AI tools and exports. The system detects personal data that arrives unmarked.
- Oversight and auditingBastion
- Every policy decision records who acted, in what role, on which object, under which rule and why, in a tamper-evident log. Compliance and audit teams can investigate past use and flag suspicious activity.
- Retention and deletionBastion
- Configurable and auditable retention policies, with verification that whatever was scheduled for deletion has truly disappeared from the system.
- Data provenanceOrbis
- Orbis preserves where each piece of data comes from and every transformation it has undergone, so that its reliability can be judged and an error can be corrected at its origin.
- Responsible AI and AI governanceAxiom
- Testing, evaluation, risk management and governance across the entire life cycle of models and agents: which model is used, with what data, for what task and with what result.
- Protection of civiliansPraetor
- In defense, Praetor integrates legal review, international humanitarian law and the rules of engagement into the planning workflow, and leaves evidence of that review. Every engagement decision requires the express and audited authorization of a person.
This section describes the design of EDELON OS. No capability is offered as a guarantee before it can be demonstrated with evidence.
Team
Everyone at Edelon answers for privacy and civil liberties. It is not the task of a department that reviews at the end: the engineering that designs a capability also answers for its effects. That responsibility includes:
- Designing, building and deploying, together with product and business, technology that protects privacy and civil liberties.
- Following the advance of technology to understand what benefits and what risks it brings for fundamental rights.
- Anticipating how privacy, data protection and AI laws change in each jurisdiction where our customers operate.
- Helping each institution handle its data and its analyses in accordance with its obligations of privacy, security, AI ethics and data integrity.
- Training everyone at Edelon to recognize a risk to rights and to resolve it as a team.
- Opening internal discussion about privacy, the ethical use of AI and new technologies.
The institutions we build for face some of the hardest questions of our time about privacy and civil liberties:
- When may a model’s inference enter a case file that affects a person?
- What must a command system record so that a civilian authority can review it afterward?
- How can information be shared between organizations without any of them losing control of what is theirs?
We answer them with their broadest consequences in mind, before writing the code.
Advisory council
Edelon will establish a Privacy and Civil Liberties Advisory Council: independent experts in privacy law, public policy, ethics and security to help us understand and resolve the complex questions we encounter in our work. The team will meet with the council regularly to review:
- Changes in privacy law, public policy and technology.
- Technical and procedural strategies to reduce risks to privacy and civil liberties.
- Opportunities to strengthen the protections in our products.
In addition to the regular meetings, the team will consult the council’s members whenever a question arises that calls for it. Their names will be published here once the council has been established.
Thinking
Beyond building technology, we care about the culture of responsibility that surrounds it, within Edelon and in dialogue with those who study it, regulate it or question it: academia, legislators, civil society and the press. That is why we publish our reasoning, so that it can be debated and refuted.
- The Vendor Is Not a Fourth BranchArchitecture of Limits, #1
- What power is actually acquired by whoever builds the system in which an institution makes its decisions, and why operating the infrastructure confers no public powers. Read the essay
- Coordinate Without ConcentratingArchitecture of Limits, #2
- How a common technological foundation can serve the three branches of government without nullifying the separation that the constitution establishes. Read the essay
AI ethics
AI ethics does not end with the model. It depends on where AI is used, for what decision, with what data and under what authority. That is why our rules cover the entire system and not just the algorithm.
- AI proposes; policy governs.
- No proposal from a model or an agent becomes an action with effects without passing through the organization’s policy and, where that policy requires it, the approval of a person with the authority to give it.
- Every proposal records its provenance.
- Which model, in which version, with what data and with what instructions, and who accepted or rejected the result: everything is recorded for later review.
- Evaluate before use and during use.
- Models and agents are tested on the real task before they enter service and monitored while in use, because performance in the laboratory does not guarantee performance in operation.
- Customer data does not train third-party models.
- We connect only models from providers that commit in writing not to retain the data, not to train on it and not to give their staff access to it.
- No third-party tracking.
- EDELON OS observability is our own: use of the platform is not sent to external tracking services.
- No autonomous lethal action.
- In defense, no engagement decision is made by a machine.
This page is complemented by our Human Rights Policy, Security Practices and Vulnerability Disclosure.