--- title: "Trust & control | How Rex keeps AI agents correct and accountable" description: "How Rex keeps its AI agents correct and under your control: supervised autonomy, your procedures, backtesting, decision traces for every action, tenant isolation, SOC 2 Type II and a no-training model policy." canonical: https://www.rex.inc/trust --- Trust & control # Agents you can audit, on rules you set. Rex runs accounts receivable work inside your finance systems. Here is how its agents stay correct, stay under your control, and keep your data secure. [Learn more](https://www.rex.inc/learn-more) [Visit the Trust Center](https://trust.rex.inc/) - SOC 2 Type II Certified - Human in the loop Approval by action type - Every action traced With the reasoning - No model training On your data, ever Control ## Autonomy you grant, one action type at a time. Rex works on a human-in-the-loop approval model. Your team keeps the policies, approval limits and decisions that need judgment, and decides how, when and for which actions independence grows. 1. Supervised **New procedures start supervised** Actions such as sending a customer email or applying a credit are surfaced for review before they run. 2. Graduated **Autonomy is earned, by action type** Approval requirements can be relaxed for specific action types as performance is proven, under policies you configure. 3. Gated **Judgment stays with your team** Work that should always need a person stays gated. Decisions outside Rex's authority are escalated with the context assembled. Correctness ## Your procedures, proven before they touch a customer. Agents follow the procedures your team already uses. Changes are tested against past work first, and finance experts at Rex oversee the work agents do. 01 Scope Agree which processes Rex will manage, how success will be measured, and which decisions stay with your team. 02 Instruct Your standard operating procedures, policies, approval rules and authority limits are encoded as agent operating procedures. 03 Validate Your dedicated Rex finance expert connects systems and validates agent performance before anything goes live. 04 Simulate Changes to a procedure are backtested against past work before they run in production. 05 Improve Agents learn from feedback and self-evaluation, each approved exception refines the procedure, and performance is reviewed against agreed KPIs and SLAs. Visibility ## Every action, with the reason behind it. Every action taken by agents and operators is recorded with an audit trail: what was done, the data used, the reasoning, the result and any escalation. Mission Control shows what every agent is working on, account by account, and Rex reports performance against the KPIs and SLAs agreed with you up front. Security & data ## Your data stays yours, inside your tenant. The full security posture, vulnerability disclosure policy and how Rex secures AI are on the [security page](https://www.rex.inc/security). **Certification** SOC 2 Type II certified, with the report, subprocessors and current posture in the [Trust Center](https://trust.rex.inc/). Adheres to ISO 27001 standards. **Tenant isolation** Each customer runs in a fully isolated environment. Data, policies and agent context are scoped per tenant, with no cross-tenant access. **Access** Enterprise SSO through WorkOS, including Okta, Azure AD and Auth0, with SCIM directory sync and role-based access control. **Encryption & residency** AES-256 encryption, in transit and at rest. GDPR compliant, with EU data residency. **Retention** Operational data is kept for the length of your contract. Inference logs roll over every 30 days, and all customer data is deleted within 30 days of a request after the contract ends. **Portal credentials** Managed in Rex or connected through a 1Password vault, with alerts when credentials expire or access fails. AI models ## One model provider, and no training on your data. AWS Bedrock operates under zero data retention by default, so prompts and completions are not stored or used for training by the model provider. ### What Rex does - Uses Anthropic's Claude models, through AWS Bedrock, for all inference - Runs inference server-side, inside Rex's infrastructure, scoped to one tenant per request - Uses your data only at inference time, to do work inside your tenant - Attributes every output to the system, and never presents it as written by a person ### What Rex doesn’t do - Train, fine-tune or host its own models - Use your data to train or fine-tune any model - Share data, context or memory across customers - Use open-source or third-party models FAQ ## Trust and control questions ### Do people approve what Rex does? Rex works on a human-in-the-loop approval model. Actions such as sending a customer email or applying a credit can be surfaced for review before they run, and every action includes a decision trace explaining why it was taken and what data informed it. Approval requirements can be relaxed over time for specific action types, but only under policies you configure. ### What stays with our team? Your team keeps the policies, approval limits, and decisions that need judgment. Rex routes disputes, sensitive accounts, and other judgment calls to the right owner with the evidence attached. You can see what was cleared, what was routed, and what still needs a person. ### Can we see what Rex did? Yes. Every action is recorded with an audit trail showing what was done, the data used, the reasoning, the result, and any escalation, for actions taken by agents and by operators. ### Which AI models does Rex use, and is our data used for training? Rex uses Anthropic's Claude models through AWS Bedrock for all inference, which runs server-side inside Rex's infrastructure. Rex does not train, fine-tune, or host its own models. Your data is never used to train or fine-tune any model and is never shared across customers. AWS Bedrock operates under zero data retention by default, so prompts and completions are not stored by the model provider. ### Is Rex secure? Rex is SOC 2 Type II certified. Each customer operates in a fully isolated environment, with data, policies, and agent context scoped to that tenant. Rex runs on AWS, and authentication is handled by WorkOS with support for enterprise SSO. The SOC 2 report, subprocessors, and current security posture are published in the Rex Trust Center. ## See it run on your accounts, under your rules. A short walkthrough of agents working accounts under your guardrails, with your team on the exceptions. The SOC 2 Type II report is available to customers on request. [Learn more](https://www.rex.inc/learn-more) [Security details](https://www.rex.inc/security)