The assurance layer for an AI enterprise

First, autonomy. Agents that hold credentials, call tools and act without a human in the path moved from demonstrations into production inside roughly two years. They are a new class of principal with a new class of failure mode, and the assurance practices around them are close to nonexistent because the practices predate them.

Second, regulation with teeth. NIS2 and DORA are in force and the EU AI Act is applying in stages, and the direction of all three is away from self-assessment towards demonstrated effectiveness. A compliance position supported by a questionnaire is becoming an exposure rather than a defence.

The question changed from whether a control is deployed to whether it still works — and almost nobody can answer the second one with evidence.

Third, board-level scepticism about attestation. Boards have watched enough incidents at organisations with complete-looking control catalogues to stop treating coverage as evidence. The question being asked internally has shifted from what is deployed to what has been proved, and the people being asked cannot currently answer.

Security validation and AI security are each already established budget lines, and the companies occupying them have raised substantially. What none of them does is cover both the AI layer and the infrastructure layer with one autonomous adversary and one evidence model, which is the specific position Veync is taking.

Veync is not a faster penetration test, a richer scenario library or a prompt-testing tool, and the product is deliberately designed so that comparison does not hold. It is a continuous adversary with a control plane and an evidence graph, covering governance, guardrails, agent authorization, AI-authored code and the deployed control estate as one surface.

Creating a category is harder than entering one, and the honest reason it is worth doing here is that the buyer's question genuinely spans both layers. A CISO asked whether the organisation's AI is safe cannot answer with a model evaluation, and a Head of AI asked the same question cannot answer with an endpoint report. Today they commission two vendors and reconcile nothing.

The wedge is narrower than the category. The fastest entry is the question no incumbent answers at all — whether autonomous agents can exceed their authority — because it is new, it is board-visible, and it has no established supplier. Control validation and framework mapping then expand the account from a position of having already proved something.

The initial market is European regulated enterprises with a named evidence obligation: financial services and insurance under DORA, essential entities under NIS2, and public-sector and healthcare deployers under the AI Act's high-risk categories. These buyers have budget, a deadline and a specific question, which is a materially easier sale than one that begins by establishing that a problem exists.

Being headquartered in Paris La Défense is a distribution decision. The density of banks, insurers and asset managers in the district means the first fifty conversations are reachable without a flight, and in this market the reference that matters is from an institution the buyer recognises as a peer under the same supervisor.

The motion is founder-led and evidence-led: a scoped first engagement that proves something specific and uncomfortable, then expansion across modules and into continuous operation. Because the output is evidence a compliance function needs on a recurring basis, the natural contract shape is a subscription rather than a project.

Expansion follows the regulation. The same obligations propagate to suppliers of in-scope entities, which opens a mid-market tier, and the EU frameworks are being used as a template in other jurisdictions, which makes the regulatory fluency portable rather than local.

This company needs three capabilities in the same room: people who have actually run offensive operations against hardened estates, people who have built and broken AI systems, and people who can sell evidence to a European compliance function. Any two without the third produces a familiar failure — a research project, a tool with no buyer, or a compliance product with nothing behind it.

Veync was founded by Anwar Ul Haq and Asad Ali Khan. Anwar brings eighteen years in enterprise security, centred on incident management and response — the work of establishing, under pressure, whether a control actually held. Asad is a network security specialist whose practice has been built inside large enterprise environments, and he brings that view of how an enterprise network really behaves to the way its defences are tested.

Both founder biographies appear on the company page. The team is being built out from that core across offensive research, platform engineering and enterprise sales, and the open roles are listed on the careers page.