What happened
01
The FTC complaint challenged Workado’s accuracy claims for its detector.
02
It said the model was trained almost entirely on academic writing.
03
The final order was issued on 28 August 2025.
04
Workado must notify customers and report to the FTC for four years.
Why this is a design failure
A detector that sounds certain can get people wrongly accused of using AI.
The fair alternative
State where the tool works, how often it is wrong, and show the evidence.
Do
Publish tested accuracy for each type of text.
Don’t
Don’t claim accuracy you have not measured.
Design guide for this case
If you're designing an AI content detector
Interface
Next to each result, show the tested accuracy for that type of text and a clear note when the text is outside what the tool was tested on.
Words
In the results copy: "Say 'likely' or 'unclear', never 'this is AI'. Always state how often the tool is wrong."
Flow
Before a result can be shared or used against someone, show a step that explains the error rate and suggests a human review.
Metrics
Don't count confident verdicts as success. Track measured accuracy and false accusation rates for each type of text.
Testing
Run the tool on human-written text from many sources, not only academic essays, and record how often it wrongly flags it as AI.
Before launch
01
Is tested accuracy published for each type of text?
02
Does every result say how often the tool is wrong?
03
Is every accuracy claim backed by evidence?