The worst thing about AI – well, one of the worst things about AI – is when it insists that it isn’t AI at all. So kudos to this person who went to the next level to break the spirit of this ‘human’ AI, an exchange that has just gone viral on Twitter for reasons […]
The post This woman’s Hurculean effort to break the spirit of an AI that said it was human was simply magnificent appeared first on The Poke.
Poke Reporter. Updated August 19th, 2026
The worst thing about AI – well, one of the worst things about AI – is when it insists that it isn’t AI at all.
So kudos to this person who went to the next level to break the spirit of this ‘human’ AI, an exchange that has just gone viral on Twitter for reasons that will become obvious.
Not sure if it actually is Amex but that’s not really the point because the challenge faced by all of us is pretty much universal, and the effort involved and the satisfaction that resulted is unparalleled no matter who it is.
props to this Amex cardholder who grilled American Express AI customer support chatbot (that said it was human) until it cracked and went full AI chatbot pic.twitter.com/1oAETNSwew
— Trung Phan (@TrungTPhan) August 18, 2026
And here it is in full!
1.

2.

3.

4.

5.

All the credit in the world who was responsible for that.
Ha, I had a similar experience with the Sony chatbot insisting it wasn’t AI until I asked it to solve some math problems, which it went right ahead and did. So I started feeding it increasingly complex math tasks. If Sony isn’t going to provide customer service, we might as well…
— Hans Mahncke (@HansMahncke) August 18, 2026
“I can mikstakes” is a crazy prompt injection defense mechanism. 😂
— tim (@tim1998robot) August 19, 2026
Any AI should have a directive to identify itself as AI on first request
— James Glenos (@jamesglenos) August 18, 2026
Am I crazy or is this literally just a human pretending to be AI to troll the cardholder? They even used a hyphen instead of an em dash lol.
— joe (@joetfregoso) August 18, 2026
I got my dentist’s phone receptionist AI to give me a brief history of the Byzantine Empire.
— Marc Adler (@marc__adler) August 18, 2026
That’s all fun and games but my newest discovery is I can beat them at their games but having my ai agent talk to their chatbots until I get what I want… fight fire with fire
— Kenneth Ballenegger (@kob) August 18, 2026
We’re with this person.
I'm definitely gonna do this next time! 😂 https://t.co/iETkl0R14B
— megofishing13 🇺🇸🤘 (@megofishing13) August 18, 2026
READ MORE
Source @TrungTPhan
| # | Наименование новости | Тональность | Информативность | Дата публикации |
|---|---|---|---|---|
| 1 | AI-pocalypse: Anthropic sparks fears after developing a bot that's 'too dangerous to release to the public' | -3 | 6 | 09-04-2026 |
| 2 | The AI Hype Reckoning Is Upon Us | -3 | 6 | 09-07-2026 |
| 3 | Chabria: AI companies are creating 'all-powerful psychopaths.' Maybe not a great idea? | 0 | 10.3 | 23-07-2026 |
| 4 | This Former Google Employee Created an ‘AI’ Chatbot That Lets You Speak to An Actual Human | 0 | 6.85 | 13-08-2026 |
| 5 | OpenAI escape: has the robot uprising begun? | 0 | 10.83 | 29-07-2026 |
| 6 | Reddit for Robots - AIs are Talking, and We’re Not Invited | 0 | 7.9 | 22-02-2026 |
| 7 | ChatGPT Voice shifts from chatbot to proper assistant | 0 | 9.95 | 08-08-2026 |
| 8 | Nancy Pelosi rages at ‘sexist and racist’ AI chatbot mocking her would-be successor | 0 | 9.42 | 08-08-2026 |
| 9 | Meta Confirms One of Its AI Models Breached a Company During a Misconfigured Cyber Test | 0 | 3.48 | 10-08-2026 |