For your role

For the operating and risk leader

The governance question about embedded AI is not whether the output is impressive. It is whether you can reconstruct, months later, what informed a decision, who approved it, and where each claim came from. That reconstruction is the design centre of the layer.

Provenance as the default

Every claim carries a source and a tier: self-reported, verified or triangulated. Every decision carries the data behind it, the reasoning, the people involved and the approvals.

That record is what turns an AI output into something a committee, an auditor or a regulator can work with. Without it, an output is a summary, and a summary cannot be defended.

Access and separation of duties

Role-based access covers operating teams, risk, compliance, external auditors and, where relevant, the partner's own clients. Each role sees what it should and no more.

Data is isolated per tenant, and residency can be selected per tenant where local rules require it.

The regulatory boundary, stated plainly

Reuben AI is software. It does not hold client money, settle transactions, act as a licensed administrator, trustee, custodian or registry, distribute securities, or provide audit, legal, tax or investment advice.

Licensing, fiduciary duties and regulatory filings stay with the partner or the fund manager. Where a page on this site refers to a regulatory requirement, it cites the primary source so your team can check it directly.

Records and retention

Retention rules are configured per tenant so the record lives as long as your obligations require. Evidence collected once is reusable rather than re-collected, which reduces both cost and the risk of inconsistent versions in circulation.

This seat, by partner archetype

By market

Questions in this seat

Can we reconstruct a decision made two years ago?

That is the intent of the provenance chain: the underlying data, the reasoning, the people and the approvals, retained under the retention rules you set.

Who is responsible if an AI output is wrong?

The decision remains the responsibility of the party making it. The layer is designed so a human reviewer can see what the output was based on before relying on it, and so the review itself is recorded.