<
https://www.techdirt.com/2026/03/23/bernie-sanders-interviewed-a-chatbot-to-expose-ais-secrets-it-has-no-secrets-it-just-agrees-with-you/>
"Senator Bernie Sanders has a viral video making the rounds in which he
“interviews” Anthropic’s Claude chatbot about the dangers of AI and privacy. It
has over two million views. Plenty of people are sharing it. And it might be
one of the most unintentionally revealing demonstrations of AI’s
actual
problems that a politician has ever produced — just not in the way Sanders
thinks.
In the video, Sanders asks Claude a series of questions about AI, privacy, and
data collection, and Claude gives a series of alarming-sounding answers about
corporate surveillance and threats to democracy. Sanders nods gravely. The
implication is clear: even the AI itself admits that AI companies are doing
terrible things to your privacy! If that doesn’t convince you, what will?
But that’s ridiculous if you actually understand how this stuff works (which
Sanders clearly does not). When you “interview” a large language model you are
talking to a very sophisticated text prediction system that is specifically
designed to give you responses that are (possibly) helpful, (hopefully)
relevant, and (obsequiously) agreeable — shaped entirely by how you framed the
question. It’s not there to help you uncover hidden truths. It’s not a
whistleblower. It’s not a witness in a congressional hearing, which is exactly
what Sanders’ staging is designed to imply.
Ask it scary questions, get scary answers. Ask it reassuring questions, get
reassuring answers. It is a
mirror, not a source.
And Sanders’ video demonstrates this — just not in the way he intended."
Cheers,
*** Xanni ***
--
mailto:xanni@xanadu.net Andrew Pam
http://xanadu.com.au/ Chief Scientist, Xanadu
https://glasswings.com.au/ Partner, Glass Wings
https://sericyb.com.au/ Manager, Serious Cybernetics
