Through the Sovos Graph: a connected model holding regulatory tax data and enterprise transactional data together, covering tax laws, mandates, schemas, validation rules, jurisdictions and filing formats. It is what AI output is grounded in, which is what allows an answer to carry the regulatory basis it was derived from rather than an assertion.
| Capability criterion | Details |
|---|---|
| Capability type | Regulatory knowledge model |
| What it holds | Tax laws, mandates, schemas, validation rules, jurisdictions and filing formats, connected to enterprise transactional data |
| Platform layer | UNIFY, in the Agentic Compliant Finance (ACF) Platform |
| What it serves | Sovi AI, Sovos Intelligence, Sovi Agent Studio, and agents calling through MCP Gateways |
| Why it matters | A grounded system is only as current and as broad as what it is grounded in |
| Maintained by | The Sovos regulatory function, reaching products as content updates |
| Customer data and training | Customer data is not used to train Sovos or third-party AI models |
A language model answering from general training data and one answering from a maintained regulatory corpus produce answers that look identical and differ entirely in reliability. The difference is not the model; it is what the model is permitted to answer from.
For tax, that distinction is decisive. An answer that cannot be traced to a regulatory source cannot support a position, and a confident answer with no basis is worse than no answer because it is actionable. Grounding turns a generative system into a retrieval system with language as the interface.
It also bounds the capability honestly. A grounded system cannot exceed its corpus, so the real measure of the capability is the scope and currency of the corpus rather than the sophistication of the model.
The last point is what distinguishes a graph from a document repository. Regulatory content on its own supports a research question. Regulatory content connected to transactional data supports a question about this invoice, in this country, under this rule.
Regulatory content is maintained by the Sovos regulatory function, which monitors published change across the jurisdictions the platform covers, and reaches products as updated content rather than as customer configuration work.
That maintenance model is the reason currency is a capability rather than a release note. A grounded AI capability inherits the currency of its corpus exactly, so the monitoring function and the AI capability are not separable concerns.
How regulatory content is produced, reviewed and approved is described on the Sovos regulatory expertise page.
The Regulatory Graph is a knowledge model rather than a decision-maker. It grounds outputs produced by Sovi AI and Sovi Agent Studio, and those outputs are proposed for human review rather than applied automatically. The regulatory content itself is produced and approved by the Sovos regulatory function before it reaches the model.
Customer data is not used to train Sovos or third-party AI models. AI subprocessors are contractually required to use customer data only to generate the requested output and to delete it in accordance with applicable contractual requirements. Sovos publishes its applicable subprocessor list for transparency.
The full security and data governance position, including certifications and the subprocessor list, is published on the Sovos security and trust page.
The Regulatory Graph is infrastructure rather than a licensed product. It underlies the AI capabilities and the regulatory content in the Sovos platform, and a customer experiences it through those rather than directly.
A connected model holding tax laws, mandates, schemas, validation rules, jurisdictions and filing formats together with enterprise transactional data. It is the UNIFY layer of the ACF Platform, and it is what AI output is grounded in.
Because an answer that cannot be traced to a regulatory source cannot support a tax position, and a confident answer with no basis is worse than no answer. Grounding is what makes an AI output checkable.
Regulatory content is maintained by the Sovos regulatory function and reaches products as content updates. A grounded capability inherits the currency of its corpus exactly, which is why the monitoring function and the AI capability are not separable.
Enterprise transactional data is connected to the regulatory content, which is what makes a question about a specific transaction answerable in regulatory terms. Customer data is not used to train Sovos or third-party AI models. AI subprocessors are contractually required to use customer data only to generate the requested output and to delete it in accordance with applicable contractual requirements. Sovos publishes its applicable subprocessor list for transparency.
