
What the analysis covered
In our analysis of 9 October 2026, we compared 393 posts about Griffin with the developer’s materials and research. One result discussed was that 26 of 54 participants in a short video conversation believed they were speaking to a human.
What limits the conclusion
A small sample and a short conversation do not prove that a system is indistinguishable from a person in every situation. It matters whether participants knew they might encounter AI, how the question was phrased and whether there was a control group. Without this context, the percentage can easily be overinterpreted.
The practical takeaway
A convincing video conversation does not, by itself, confirm the other person’s identity. For consequential decisions, a separate verification channel is useful: a callback to a previously known number or an agreed confirmation procedure. The article’s data describes the experiment as of the original analysis date.
Adapted from the author’s channel. Original analysis and source links.


