The principle
We run AI and people on the same queues inside regulated businesses. Our security model follows from that: AI systems draft and filter, a named person approves anything that reaches a customer or moves money, and every action is logged. The controls below are a summary. Full detail, including policies and evidence, is available to clients and prospects under a non-disclosure agreement.
Access control
- Role-based access with least privilege. Staff see only the clients, queues and data their role requires.
- Multi-factor authentication on all systems that hold client data.
- Client environments are isolated from each other. Where a client prefers, our people work inside the client's own systems and nothing leaves them.
- Access is granted and removed through a joiner, mover and leaver process, with periodic access reviews.
Data handling
- Data in transit is encrypted. Data at rest is encrypted on the systems we operate.
- We collect and keep the minimum needed for the work, for the period agreed in the contract.
- Client data is not used to train models shared with other clients.
- Personal data is scrubbed or minimised before it enters any retrieval corpus or embedding store.
AI-specific controls
- Draft-approve. AI output that reaches a customer or moves money is held in an approval queue until a named person approves, edits or rejects it.
- Guardrails. Suitability, frequency, timing and regulatory-hold rules stop inappropriate recommendations from being generated.
- Audit log. Inputs, outputs, overrides, reviewer and timestamp are recorded so any decision can be replayed.
- Model inventory. Every model in production has an owner, a version, a validation record and monitoring with escalation thresholds.
- Third-party models. External model providers and APIs are treated as governed dependencies. No client data is sent to a public AI service without the client's written agreement on which service, which data and under what terms.
People
- Staff sign confidentiality agreements and complete security and data-protection training before they work on client queues.
- Background checks are carried out where the client's sector or the role requires them.
Incidents
If we become aware of a security incident affecting a client's data, we notify the client without undue delay and within the timelines set in the contract, share what we know, and work with the client on containment and any regulatory notification.
Reporting a vulnerability
If you believe you have found a security issue on platorix.com or in a system we operate, email contact@platorix.com with "Security" in the subject line. Please do not access or alter data that is not yours. We will acknowledge your report and keep you informed while we investigate.