1. markst's role
Under the Act, obligations differ between providers of AI systems, deployers (users) of AI systems, and providers of general-purpose AI (GPAI) models. markst is:
- Not a provider of an AI system within the meaning of Article 3(1), because the Verus valuation engine is deterministic, has no learned weights in its pricing path, and is not an AI system as defined in Article 3(1). See /methodology and /legal/ai-usage.
- A deployer of a general-purpose AI model (Anthropic Claude) when optional AI-assisted extraction or summarisation features are enabled by a customer. In that capacity, markst applies Article 26 obligations proportionately (transparency to affected persons, human-oversight, logging).
- Not a GPAI model provider — markst does not train or distribute a foundation model.
2. Why Verus is not a high-risk AI system
Annex III of the Act lists the domains that render a system high-risk. The item closest to private credit is Annex III §5(b) — AI systems intended to be used to evaluate the creditworthiness of natural persons or to establish their credit score. Verus evaluates the fair value of credit facilities extended to legal entities(companies, funds, SPVs). Natural-person credit scoring is out of scope.
Other Annex III items (law enforcement, biometrics, migration, judicial, employment, education, critical infrastructure) are clearly not engaged. Article 6(2) does not capture Verus.
The pricing engine itself is deterministic and contains no learned weights, so even if an Annex III category were read to include corporate credit valuation, the engine would not meet the Article 3(1) definition of an “AI system”.
3. GPAI consumption disclosure
When a customer enables AI-assisted extraction or summarisation, Verus calls the Anthropic Claude Enterprise API. markst discloses this use to affected persons via /legal/ai-usage and logs the model version, prompt category, and parsed output. Under Article 26(7), any generated text is identifiable as AI-generated content to the end-user; in practice, AI output in Verus appears only as a structured draft that a human confirms before it is committed to an input.
The Anthropic Enterprise API terms prohibit training on customer content and apply a short retention window. markst does not itself use customer content to train any model.
4. Prohibited practices
Article 5 practices (manipulative, exploitative, social scoring, untargeted facial-image scraping, real-time biometric identification, etc.) are out of scope for the Verus use case and are neither supported nor enabled by any part of the platform.
5. Transparency obligations (Article 50)
Where a customer uses the Verus's optional AI summarisation feature to generate human-facing text (for example an executive summary of a valuation), the resulting output is labelled as AI-generated in the UI and in exports where it is reproduced. A deterministic report is always produced first; any AI-generated prose sits alongside and is clearly identified.
6. Logging and record-keeping
Each AI-assisted call records the model vendor and version, the prompt category (extraction vs summarisation), the parsed output, and the timestamp. These logs are retained alongside the associated run record for the life of the contract plus statutory retention.
7. Opt-out
Customers may disable AI features per tenant. The deterministic pipeline is unaffected. Contact support@markst.com to provision an AI-off environment.
8. Standards alignment
- ISO/IEC 42001 (AI management systems) — we intend to align our AI-governance controls to 42001 in parallel with the ISO 27001 engagement (see /trust §2).
- NIST AI RMF 1.0 — informally consulted for risk categorisation of the extraction and summarisation features.
9. Conformity assessment position
Because Verus is neither a high-risk AI system nor a GPAI model in the Act's sense, no Article 43 conformity assessment or CE marking applies to it. markst's obligations as a deployer of a GPAI model are operational (Article 26) and documentary (this page). If a future regulation change or a client-specific deployment placed the system in-scope of Article 6(2), we would engage the relevant notified body proactively.
10. Contact
AI-governance queries: ai@markst.com. Counsel correspondence: legal@markst.com.