Responsible AI is the product, not a page.
Every commitment below is enforced by the CAI Score framework our agents ship with, not promised in marketing copy. Here is what we hold ourselves to, and how you can verify it.
Human accountability
A named human role is built into every CAI tier. Higher-stakes agents assist and evidence decisions; a person stays accountable for every call.
Bounded autonomy
Agents operate strictly within their certified tier. No agent in our catalogue runs fully autonomously today, L5 exists in the framework and is deliberately unused.
Your data stays yours
Agents run in your own Microsoft tenant. We do not process your business data, and it is never used to train models.
Evidence by default
Every agent action is logged, time-stamped and attributable. Audit answers exist before anyone asks the question.
Regulatory alignment
The CAI Score is designed around EU AI Act, DORA and ISO/IEC 42001 obligations: risk classification, oversight, logging, transparency. It supports your compliance work; it does not replace legal review.
Open methodology
Tier definitions, human roles and review periods are documented and challengeable. If you disagree with a classification, we want to hear it.
Reporting concerns
Found a security issue, a misclassified agent, or behaviour that does not match these commitments? Write to hello@colleagueai.ai or use our security.txt. We acknowledge reports within 5 business days.