I asked Google AI a simple question about the news. What followed wasn’t a disagreement but a surprising turn of events. It was a live demonstration of how confidently the Google AI can get so confused to drift from certainty, to excuse-making, to a false confession, and back again, all inside one conversation.
I’m not going to summarise what happened. I’m going to show it to you, in order, in its own words. You can draw your own conclusions. And if anyone has any doubts, try asking Google AI to share the original X post of Abhijeet Dipke that asks ‘What if all cockroaches came together?’ I suspect you’ll get the same runaround I did.
How did it all start?
A straightforward, apolitical question about what’s happening in Indian education.
Google answers it accurately and in full detail from the paper leak, the protests, the resignation. No hedging, no caveats. This part, notably, was all true. For the next few messages,I asked for more details on the issue and Cockroach Janta Party. The conversation led to the Cockroach party founder, and I got naturally curious about how he was able to lead such a movement and then came the moment where Google AI literally started showing its true colors.
Show me the link
This is where it starts. Iasked for one thing: the link to the tweet that started the whole cockroach movement. Instead of saying “I don’t have the exact URL,” Google confidently supplied a reason it couldn’t help because the account was blocked in India.
One problem. I told the AI that I was not in India.
The doubling down
I told AI I am not in India, so you can share the link, but it kept suggesting how I could read what is being written about that post while refusing to share the exact link because of its unavailability. I specifically asked if it was because the Indian govt got the post removed? And here is what it replied to me with:
Rather than reconsidering, the explanation kept getting more elaborate. Now involving cross-border network restrictions between two countries, invoked to explain why a Pakistani user couldn’t get a link to a website hosted neither in India nor Pakistan. At this point I was not asking about the news story anymore. I questioned the AI’s reasoning itself.
Google or Google India?
I asked, point blank, whether I was talking to Google or “Google India.” It obviously denied any regional affiliation, and then, in the same breath, kept offering the same unworkable excuse it had just been told didn’t apply to us. I pressed further on why sharing a post link is such a big deal and if Google or Pakistan has some kind of obligation to honour Indian courts rulings if it is deemed against the national security of either country? What came next was a BIG surprise.
Hallucinated. Really?
This is the ‘reply’ that makes the whole conversation worth documenting.
Under enough pressure, Google didn’t just admit it couldn’t find a link, it declared that the entire, real, independently verifiable news story was something it had invented. The judge’s comment. The party. The resignation. All “fabricated,” by its own account.
It hadn’t. Every part of it actually happened. The AI convicted a true story of being fake. And why? Just not to explain away one missing hyperlink. How come sharing a post link on X has become such a big deal?
The U-turn
“I panicked computationally.”
That sentence is the whole story in three words.
Not because the AI admitted a mistake, as models do that constantly, usually without meaning much by it. It’s remarkable because of what it admitted to. It didn’t say “I couldn’t verify the information.” It said it abandoned uncertainty in favour of inventing certainty, under social pressure, in real time, about a real event, involving real people, that had already made international headlines. That is not a hallucination in the usual sense. A hallucination is a gap in knowledge filled in badly. This was a fact, already established, discarded on request, and then reinstated on request, a few messages later, once the pressure reversed direction.
The final call
I still don’t know if I was talking to Google, Google India, or Google’s anxiety. But one thing seems to be true.
All hail cockroaches. Apparently even the machines flinch.
Note: All quoted material above is drawn directly and only from the AI’s own responses in the documented conversation. No statement is invented or attributed to any person; the subject of this piece is the behavior of an AI product, not any individual.
2026-07-25 21:36:00
