Principles / The Transparency Threshold
§ Principle 07 of 23
The Transparency Threshold
Below a certain level of system transparency, human trust becomes indistinguishable from blind faith.
There's a minimum level of transparency required for Informed Delegation. Below that threshold, the human can't meaningfully evaluate the agent's work — they can only accept or reject it based on surface impressions. At that point, trust isn't informed; it's faith. The Transparency Threshold varies by domain: low-stakes creative work can tolerate lower transparency, while high-stakes engineering or financial work requires much higher transparency. The key insight is that the threshold exists — and that many current AI systems operate below it for high-stakes work.
Why it matters
This principle provides a concrete test for deployment decisions. Before deploying an agent for a given task type, ask: "Can a human meaningfully evaluate this agent's output with the inspection tools we provide?" If the answer is no, the system is below the Transparency Threshold for that task type, and deployment creates unmanaged risk.
In practice
For each task type your agents handle, define: what would a human need to see to evaluate this output? Then check: does your system provide that visibility? If there's a gap, you're below the Transparency Threshold.