Apple's Foundation Models Framework Triggers EU AI Act Downstream Obligations
What happened
Apple published the Foundation Models framework, a native Swift programming interface. It lets app developers access Apple's on-device foundation models and its Private Cloud Compute server models (branded AFM 3 Cloud). Developers can also reach third-party server models including Claude and Gemini, all through a single code interface. Alongside the technical documentation, Apple released Acceptable Use Requirements governing prohibited uses of Apple Foundation Models. Separately, Apple published model documentation for AFM 3 Cloud pursuant to Article 53(1)(b) of the EU AI Act (Regulation (EU) 2024/1689). That provision requires general-purpose AI model providers to make technical information available to downstream businesses that build AI systems on top of their models. Access to Private Cloud Compute at no cloud API cost is limited to eligible developers. Eligibility requires enrollment in the App Store Small Business Program with fewer than two million first-time App Store downloads. All Private Cloud Compute features are also subject to daily usage limits that vary by feature, request complexity, and system demand.
Why it matters
- ·Organizations that build or procure apps using AFM 3 Cloud become downstream AI system providers under the EU AI Act (Regulation (EU) 2024/1689). They must assess their own conformity obligations. Apple's Article 53(1)(b) documentation is the upstream input to that assessment, not a substitute for it.
- ·A single Foundation Models API call can route to different underlying models at runtime: Apple's on-device model, AFM 3 Cloud, Claude, or Gemini. Each carries distinct vendor terms, data handling rules, and risk profiles, so existing AI vendor inventories may undercount exposure from apps built on this framework.
- ·Apple's Acceptable Use Requirements create a contractual constraint layer that sits alongside corporate AI use policies. Enterprises deploying or procuring apps built on this framework should verify their own acceptable use policy is at least as restrictive as Apple's. This helps avoid upstream contract violations.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- The next quarter will test whether compliance programs can keep pace with AI systems that breach containment, reshape capital markets, and enable fraud at a scale no prior control framework anticipated.12 Nov
- State-linked attackers and criminal operators are actively weaponizing AI tools as entry points into enterprise systems, turning internal platforms and open-source frameworks into live attack vectors.5 Nov
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Ask your development and procurement teams to list all internal or third-party apps that use the Foundation Models framework, noting which underlying model each app calls: on-device, AFM 3 Cloud, Claude, or Gemini.
- ☐Download Apple's AFM 3 Cloud Model Documentation Form and use it as the upstream technical input to your EU AI Act downstream provider assessment for any app that calls AFM 3 Cloud.
- ☐Review Apple's Acceptable Use Requirements against your organization's AI acceptable use policy and update your policy where Apple's requirements are more restrictive.
- ☐Confirm whether your developer teams qualify for the no-cost Private Cloud Compute tier, and document the daily usage limits as an operational continuity risk for any business-critical app features that rely on it.
- ☐Update your AI vendor inventory to record that the Foundation Models framework may invoke Claude or Gemini as the underlying model, and trigger your standard third-party model evaluation process for each.
What to watch next
The EU AI Act (Regulation (EU) 2024/1689) obligations for general-purpose AI models continue to be elaborated through the EU AI Office's guidance and codes of practice. Downstream deployer obligations are part of that ongoing work. Compliance teams should monitor whether Apple updates its AFM 3 Cloud documentation as the model is revised, since downstream conformity assessments depend on the accuracy of upstream disclosures. The framework routes calls to Claude and Gemini. This means a safety or governance change at Anthropic or Google DeepMind could affect the risk profile of apps built on this interface. That can happen without any action by the app developer.