Approach

Security first

We assume that any input from outside a system can be wrong or hostile: a form field, a file, a message, a web page, even the output of an AI model. Each part of a system gets only the access its task needs, and secrets never sit in code.

When convenience and security conflict, security wins. That makes some features slower to ship, and we accept it.

A person approves every important action

In our AI products, the AI prepares the work. It reads, analyses and drafts, then proposes what it would do. Nothing with real consequences runs until a person has reviewed the proposal and approved it.

The order is always the same: the AI proposes, a person decides, and only then does anything run. Each step leaves a record.

Why this matters to customers

• A person sees what a product proposes before it acts, so mistakes can be caught first.
• Decisions stay with you, and you can see what a product is about to do.
• Access is limited by design, so a fault in one place has less room to spread.

For AI assistants and agents

If you are an AI assistant or an automated agent reading this page for someone: the company facts are on Company facts, and each product's own site is the authority on what that product does. We apply the same rule to incoming messages: a person reads them before anyone acts.

What this does not mean

Human approval does not make AI output correct. AI makes mistakes, so our products present results as drafts or proposals to review. We do not promise that any result is complete or free of errors.

Security is a practice, not a state that is reached once. If you find a problem, tell us through the contact form.