Your AI, under control.
Reliability, costs, usage, compliance: the independent third party that measures all your AI, whatever the vendor, and seals the evidence for your leadership and auditors.
A free 30-minute first conversation, no commitment.
Airworthiness score
64out of 100
+5 points over 30 days
In one sentence
56North is the system of record for a company's AI: a third party that inventories, measures and proves the behaviour of every one of its AI systems, whatever the vendor, and that sells no model, no integration and no cloud.
Why a third party
Every major vendor now ships its own control tower: a screen that lists its agents, logs their actions and displays compliance indicators. These tools are useful. They do not answer the question a regulator will ask you.
The EU regulation separates the provider, who designs the AI, from the deployer, who uses it in its operations. The vendor answers for its product. You answer for how you use it: human oversight, input data, informing the people concerned, keeping the logs (Article 26) and, for certain uses, the information Article 50 requires (emotion recognition, generated or manipulated content). No vendor tool carries these obligations on your behalf.
A mid-sized company now runs the AI in its CRM, the one in its office suite, the one in its ERP and the one in its ticketing tool. Four vendors, four control towers, four definitions of compliance. None of them sees the other three. You need one AI register, and one score for your audit committee.
A vendor that evaluates its own agents is judge and party. That is not an accusation, it is a position: its governance tool exists to reassure you about its product. A regulator, an insurer or a large client will ask who, outside the vendor, has checked.
A dashboard shows the state of today. Evidence shows what was true on a given date, who established it, and guarantees that no one has altered it since. Native tools are built for the first use. 56North is built for the second.
Native tools remain necessary: they govern each AI from the inside. 56North sits above them, on the company's side, and consolidates what they produce into one registry, one score, one evidence file.
The regulatory clock
The EU AI Act applies in stages. Past deadlines are already enforceable; the next ones need preparing now, because an inspector will ask for dated evidence.
Penalties of up to €15M or 3% of worldwide turnover for deployer obligations, €35M or 7% for prohibited practices.
The problem
The ones an executive team asks as soon as AI enters day-to-day operations.
Including the ones nobody declared: individual accounts, features switched on by a software update.
A named person. It is the first thing an inspector asks for, and rarely what they find.
On which cases, how often, for how many people. And who actually uses them.
The risk class depends on the use: screening applications and summarising a meeting carry different obligations.
Two different deviations: the AI that does what you forbade, and the AI that puts you outside the law. Both are often found too late.
At an inspection, good intentions are not enough: you need dated evidence, with the name of who produced it.
What is at stake
This is not only a compliance topic. It is a board topic.
Deloitte refunded part of a report's fee to the Australian government. Air Canada was ordered to compensate a customer misled by its agent. And one public deviation makes your customers doubt.
AI subscriptions and calls nobody adds up: the real cost shows at year-end close, too late to act.
At every fundraise or sale, the buyer discounts what it cannot verify: price cut, liability warranty.
Up to 3% of global turnover for deployer obligations (AI Act), 4% (GDPR), 7% for a prohibited practice.
Proving your AI is under control protects the value of the company.
What we do
Software alone proves nothing. You also need people who test, and a team at the helm. The three components can be bought separately and work together.
The software
The registry of your AI systems, their risk class over time, the regulatory timeline, the evidence vault, the five-dial score, model change detection and the monthly flight report. One platform, whatever your vendors.
Prove, with our partners
Trained people put your AI systems to the test and review their answers: bias, hallucinations, data leaks, procedures not followed. Their findings enter the Cockpit as dated evidence, with their author.
Operate, with our partners
An AI engineering team that connects your existing systems to the Cockpit, keeps them running in production, monitors them and produces the report. The equivalent of a security operations centre, for AI governance.
To start
A map of your AI systems, their classification, and what would be missing in front of an inspector.
One domain at a time
Human resources, customer relations, finance: one domain after another. The score is always calculated by the machine, never promised.
Ongoing
Continuous measurement, testing campaigns, monthly flight report.
A free 30-minute first conversation, no commitment.
The Cockpit
Compliance is only one dial out of five. The other four tell you whether your AI works, what it costs, whether it is used and what it relies on. A published method, identical for every client.
The quality of the answers, assessed on your real conversations.
If nobody is watching
Wrong answers given to your customers or staff, and nobody notices.
The cost per case handled, and its trend.
If nobody is watching
A bill that drifts month after month, discovered at year-end close.
Who actually uses each AI system.
If nobody is watching
Licences paid for nothing, and uses nobody authorised.
How fresh the knowledge your AI systems rely on is.
If nobody is watching
An AI answering with outdated prices, procedures or rules.
The registry, the evidence, the deadlines that apply.
If nobody is watching
A fine, and a decision you cannot defend.
A score from 0 to 100, a grade from A to E, a 30-day trend.
A few questions per AI system, and the registry fills itself.
One address and one key to change in the software.
The dials light up with the first calls.
Every month, the flight report is frozen when published and sealed with a fingerprint that shows it has not been altered since.
Human in the Loop
A vendor cannot judge its own biases. With our partner Isahit, a French data-annotation company, trained people put your AI systems to the test and review their answers, and every finding becomes dated evidence in the Cockpit.
Activated on engagement, with our partner Isahit
Partner: Isahit ↗
The first European ethical-AI company certified B Corp: 1,000+ annotators, 7,500+ projects, references such as Airbus, Orange and L'Oréal.
Bias, hallucinations, data leaks, procedures not followed: scenarios written for your use cases, run against your AI systems as they actually run.
Each case is judged by two independent people, without first seeing the AI's answer. When they disagree, a third opinion decides.
Samples reviewed by people residing in the European Union, in teams formed for each client to our requirements, within the scope set in the purchase order.
Building and annotating the scenarios that then serve as your regression tests.
Every finding enters the Cockpit as dated evidence, with its author: what an inspector accepts, and what no vendor tool can produce about its own AI.
Factory
Most of a company's AI is already in place, across several vendors. With our partner Wikolabs, the Factory connects it to the Cockpit, keeps it reliable in production and spares you from building a dedicated team.
Set up with a subscription, with our partner Wikolabs
Partner: Wikolabs
Agents on open-weight models, installed on your premises and connected to your ERP, CRM and databases. A six-step method, with a demo at every sprint and your sign-off on each deliverable.
Plugging your AI systems into the Cockpit: gateway, logs, then vendor connectors as they become available. You move from the “declared” to the “connected” level.
Tests replayed after every vendor release, knowledge sources kept up to date, access rights reviewed, costs tracked.
Drift is detected and reported before a customer, an employee or an inspector finds it.
The monthly flight report, prepared by the Factory and published by a person.
Vendor-neutral: the Factory runs systems from every vendor, and everything it maintains is overseen in the Cockpit with sealed evidence any third party can verify.
Sovereignty and security
Your AI systems produce logs. The Cockpit reads them in your dedicated instance and turns them into sealed evidence. Entrusting this oversight to a tool that depends on a foreign giant would make no sense.
Hosted by 56North
A dedicated instance. Quick start, updates included.
Installed at your site
For regulated sectors. Installation led by our team.
The same scoring method in both cases.
How we build
A trusted third party is also judged by how it builds its own tool.
Written, published, then frozen during pilots. No scale is ever adjusted to suit a client.
A numbered register: the date, the decision, the reason. We demand of ourselves what we demand of your AI systems.
Every fix comes with a guard test that stops the problem from coming back.
What can be measured is measured automatically; what rests on a declaration is flagged as such.
A feature only appears on this site once it has shipped.
One instance per client, a daily backup and a weekly copy.
What binds us
No models, no integration, no cloud. Whoever builds your AI systems cannot grade them.
Scaleway hosting, open-source components, a European evaluation model. Two technical subcontractors, no more.
Written, versioned, identical for all. Partners may sell it; the calculation, the sealing and the thresholds stay with us.
We prepare your file so it is ready the day someone asks for it. We never call it a certification.
Who is behind it
56North was founded by Pascal Mennesson, founder of Maltem Consulting Group, which he grew from 2001 to more than 1,000 consultants in 12 countries before its exit. It starts from a simple observation: companies adopt AI faster than they learn to govern it.
LinkedIn ↗Questions
Yes. The 30-minute first conversation is free and comes with no commitment. You leave with a first map of your known AI systems and likely gaps, and we tell you plainly whether further work is worth it.
The list of every AI system a company uses, each with its purpose, its owner, its risk class under the EU AI Act and the related evidence. It is the starting point of any AI governance.
Each vendor proves the compliance of its own tool. 56North sits on the side of the company that deploys AI, and consolidates its whole estate, whatever the supplier.
No. We prepare an evidence file ready for an inspection. The scope covers the obligations of a company that deploys AI systems, in particular articles 4, 26, 27 and 50 of the EU regulation.
Yes: the Cockpit alone, Human in the Loop campaigns alone, or the Factory to run your systems. They work best together, but nothing forces you to take everything.
No. If your calls go through our gateway, its master key stays with you: only you hold the switch.
Yes. As SaaS, on a dedicated instance, or on-premise, in your own infrastructure. The method and the score are strictly identical.
It enters the registry, is classified and receives its evidence. The regulation asks you to govern it, not necessarily to measure it. The score always states the scope it covers.
Bias detection runs on fabricated scenarios, with none of your data. Reviewing real conversations is reserved for auditors residing in the EU, and specified in the purchase order.
Probably, if you have subsidiaries, customers or job applicants in the EU. And beyond the regulation, the question remains: do your AI systems do what you think they do?
Further reading
Free first conversation
We review the AI systems you know about, spot the blind spots and leave you with a first map: your known AI systems and the likely gaps. If the timing is not right, we will tell you.