In this article7 sections
The dual-role trap most SaaS companies walk straight into
EU AI Act SaaS compliance is more complicated than it looks for one simple reason: most SaaS companies are not just one thing under the law. You build features on top of foundation models or AI APIs and ship them under your own name: that makes you a provider. You also run internal operations using AI tools from third-party vendors: that makes you a deployer. The AI Act treats those roles differently, and it assigns separate obligations to each. Miss one side of the equation and you have a gap.
This is not a hypothetical edge case. A CRM platform that adds an AI-powered lead-scoring feature is a provider of that feature to its customers. The same company probably uses an AI-assisted HR tool to screen job applicants. Under Article 3 of the AI Act, that makes them both a provider and a deployer, simultaneously, in the same working week.
What the AI Act actually means by provider and deployer
The definitions matter, so here they are plainly.
A provider is any natural or legal person that develops an AI system or general-purpose AI model, or has one developed, and places it on the market or puts it into service under its own name or trademark. If you build an AI feature and offer it to your customers under your brand, you are the provider of that AI system, even if the underlying model belongs to OpenAI or Anthropic and you are just wrapping it. The model maker remains the provider of the general-purpose AI model.
A deployer is any natural or legal person that uses an AI system under its authority, except for personal, non-professional use. If you buy a SaaS HR tool with an AI screening module and you use it to filter CVs, you are the deployer. The HR tool vendor is the provider.
Article 25 is the article most SaaS teams have not read yet. It deals with responsibilities along the value chain for high-risk AI systems. A distributor, importer, deployer or other third party becomes the provider of a high-risk AI system if it puts its name or trademark on a high-risk system already on the market, makes a substantial modification to a high-risk system so that it remains high-risk, or changes the intended purpose of an AI system, including a general-purpose AI system, so that it becomes high-risk.
In plain terms: if you take a general-purpose API and build a product whose purpose is, say, ranking job applicants or scoring creditworthiness, you are the provider of a high-risk AI system, with the full provider obligation set. If your product does something that is not high-risk, Article 25 does not change your role, but you are still the provider of your own AI feature.
The three-layer stack you need to map
Think of your AI exposure in three layers.
Layer 1 — Upstream vendors you rely on. This is your AI stack as a deployer. AI-assisted analytics tools, automated HR platforms, AI customer service widgets you have licensed and use internally. For each of these, you are the deployer and the original vendor is the provider. For all of them, Article 4 applies: take measures to support the AI literacy of the staff who use them. For high-risk systems, Article 26 adds deployer obligations from 2 December 2027 (Annex III): use the system according to the instructions, keep the logs it generates for at least six months, assign human oversight to competent people, and inform workers and affected individuals where required. A fundamental rights impact assessment (Article 27) is only required of public bodies, private providers of public services, and deployers of credit-scoring or life and health insurance pricing AI.
Layer 2 — AI features you have built. This is your AI stack as a provider. Anything your engineering or product team has built that qualifies as an AI system under the definition in Article 3(1) and ships to your customers. For high-risk applications, such as the recruitment tools and creditworthiness scoring listed in Annex III or AI safety components of products covered by Annex I, you face the full Article 16 provider obligation set: conformity assessment, technical documentation, quality management system, EU declaration of conformity, registration, post-market monitoring. SMEs and small mid-caps can provide technical documentation in a simplified form (Article 11).
Layer 3 — The grey zone. This is where most SaaS companies actually live. You take a foundation model API, build a product feature around it, and sell that to business customers. You are not the original model developer. But you have made design decisions about how the model is used, what data it sees, what outputs it produces, and what your customers do with those outputs. The Act does not let you stay neutral here. You are the provider of your own AI feature. And if your feature gives a general-purpose model a high-risk purpose, you are the provider of a high-risk system.
Where the obligations multiply
The practical burden is not that any one obligation is crushing. The burden is that you are managing obligations from two directions at once.
As a provider, the transparency duties in Article 50 have applied since 2 August 2026. The obligations for high-risk AI systems apply from 2 December 2027 for Annex III systems and 2 August 2028 for AI in products covered by Annex I, after the Digital Omnibus on AI (Regulation (EU) 2026/1744) moved them. From then, you need technical documentation ready, a quality management system in place, and conformity assessment completed before a high-risk product reaches EU customers.
As a deployer, some obligations are already live. The prohibitions under Article 5, covering unacceptable-risk AI practices such as social scoring, workplace emotion recognition and certain biometric categorisation, have applied since 2 February 2025, and they cover the use of such systems, not only their sale. Two more prohibitions, on non-consensual sexual deepfakes and AI-generated child sexual abuse material, apply from 2 December 2026. If a system in your stack is used for a prohibited practice, using it is itself a breach.
AI literacy obligations under Article 4 have also applied since 2 February 2025. Since the Omnibus, you must take measures to support the development of AI literacy of the staff who work with AI systems; no specific level per person is required. Proportionate training, with a record of it as your evidence, is the practical answer. That applies whether those systems are your own product or a third-party tool.
Penalties under Article 99 reach €35 million or 7% of global annual turnover for prohibited practices under Article 5. Breaches of provider and deployer obligations, including Article 50 transparency, carry up to €15 million or 3%. For SMEs and start-ups, the lower of the two figures in each tier is the maximum; small mid-caps get the lower figure for the €15 million and €7.5 million tiers.
Allocating responsibility in your contracts
One of the most practical things a SaaS company can do right now is audit its vendor contracts and its customer terms.
On the vendor side: providers of high-risk systems must give deployers instructions for use (Article 13), and general-purpose model providers must give the providers who build on their models the information they need to understand the model and meet their own obligations (Article 53(1)(b)). If you build a high-risk system with components or tools from a third-party supplier, Article 25 sets cooperation and information duties along that value chain, so make sure your contract secures the information and assistance you need. If your AI API contract does not mention the AI Act at all, ask: what limitations apply, what documentation is available, and what logs can you get?
On the customer side: Article 50 requires you, as provider, to design AI systems that interact directly with people so that those people are informed they are dealing with AI, unless that is obvious, and to mark synthetic audio, image, video or text output in a machine-readable way. Generative systems already on the market before 2 August 2026 have until 2 December 2026 for that marking. If your feature is high-risk, your customers also need your instructions for use.
A short AI Act statement, attached to your product documentation or embedded in your terms, is a good habit even where the law does not require it. It can state the risk classification of your AI features, the data inputs used, the human oversight mechanism available, and who to contact for complaints.
Building a lean compliance structure that covers both roles
For a SaaS company with 20 to 200 employees, a full compliance programme does not need to be a six-month project. It needs to be a structured inventory.
Start with an AI system register. List every AI system your company builds or uses. For each one, record: is this system a provider or deployer relationship? What risk category does it fall into? What obligations apply? Who internally owns compliance for it?
For your provider-side systems, the next step is a preliminary risk classification against Annex III. Many SaaS AI features will fall into limited-risk or minimal-risk categories, which means transparency duties at most rather than full conformity assessment. Know your classification now, so a high-risk feature has a plan well before 2 December 2027.
For your deployer-side systems, review each vendor contract at the next renewal. Flag any gaps in documentation and instructions. For any system that touches employment, credit, education, or safety, check whether it is high-risk and prepare the Article 26 measures, and check whether you are in the small group that needs a fundamental rights impact assessment.
Document what you do. A supervisor that asks questions will want to see evidence: training records, risk classifications, oversight procedures, incident handling. For Article 4 there is no legal record-keeping duty, but records are how you show your measures. For high-risk providers, documentation is a legal requirement.
Knowing whether you are a provider, a deployer, or both is the first question the AI Act asks of every SaaS company. Answer it clearly, map your stack against the right obligation sets, and you have a compliance programme that is proportionate and defensible. Start with your AI system register this week. It takes an afternoon and it gives you the foundation for everything else.
Run a free 2-minute compliance check at comply.khairos.ai to see which obligations apply to your specific stack right now.
References
Part of our guide to Audit, Evidence & Governance
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.