AI Safety & Data Governance
Every Gerento agent operates under four standing constraints: read-only access to source systems, a human approval step before any output leaves the organisation, no authority to approve spend or commit the company, and a full audit trail of every action and the data behind it. Agents draft, track, route and recommend. They do not decide, publish or transact.
Updated
1. What an agent is allowed to do
A Gerento agent reads data, produces drafts and reports, tracks open work, and routes items to named people. It operates continuously on a defined scope of work for one role.
2. What an agent is never allowed to do
- Write to source systems. Access is read-only. Agents do not edit CRM records, change forecasts, alter specifications or modify warehouse tables.
- Publish externally. Nothing reaches a customer, retailer, agency, social account or storefront without a person approving it.
- Approve or commit. Agents hold no spend authority, no approval rights in a regulatory or legal workflow, and cannot commit the company to anything.
- Make the judgment call. Where a decision requires weighing relationships, politics or risk, the agent escalates with evidence instead of deciding.
3. Human in the loop, by design
Each agent role page publishes an explicit list of what that agent cannot do, and the demonstrations on this site keep the same boundary: the agent scores, drafts and routes, and a person accepts or rejects. The boundary is a product constraint, not a setting a customer can turn off.
4. Data handling
- Customer data is processed for the agent's defined scope of work and is not used to train third-party models.
- Access is scoped to the systems needed for that role, at the minimum permission level that makes the work possible.
- Market and consumer signals come from the Tastewise data platform; company-specific data stays within the customer's environment and scope.
5. Auditability
Every agent action is recorded with a timestamp, the skill that ran, and the data it read. Outputs cite their source and method so a number in a report can be reproduced and challenged rather than taken on trust. The activity feed shown in the product demonstrations is the same mechanism.
6. Accuracy and its limits
Agents can be wrong. Outputs carry their sources so an error is traceable, base sizes and assumptions are stated rather than hidden, and anything outside an expected range is flagged rather than smoothed over. Where available data cannot answer a question, the agent reports that instead of asserting an answer.
7. Employment decisions
Gerento agents are not used to screen, rank, score or reject job candidates, and the job index on this site is not an applicant tracking system. Nothing published here should be treated as an employment decision or a recommendation about any individual.
8. Reporting a concern
To report an inaccurate output, a governance concern or a suspected misuse, email support@gerento.ai. See also our Privacy Policy and Terms of Service.