Watched Gemini catch a breaking $280M AAVE exploit mid-conversation, retract it under pressure, then confirm it was real – all while I was debating entry points

This is a wild one. I was in the middle of a live high-leverage futures analysis session with Gemini Pro on AAVE when something unprecedented happened in real time.

Full Gemini chat here

The Setup

AAVE had just printed a violent, unexpected 7–9% red candle. Nothing in the news, reddit or elsewhere. No obvious catalyst. I was suspicious (I've been trading crypto over a decade) and with time you notice when something is not right. This was already an hour past the initial spike and still no real info, so I decided to try to ask AI to look into this, let the google mastermind search thru every online source right? It gave me a very bullish summary and that this is just normal market movement more or less.
But after I saw it talk about wrong entry price points, I sent a screenshot and suspected it's not tracking real time data. Still, gemini sticked to his guns that there's nothing fishy going on, even with me questioning every single thing it literally said this:

The "Insider Info" Suspicion.
I just scanned the latest crypto news and governance boards, and there are absolutely zero indications of an exploit, hack, or insider dump.

So I decided to trust it and was debating futures entry points and strategy. Discussing stop-hunt mechanics, breakouts, margin efficiency, the whole thing. This was actually the first time I was using AI seriously to discuss futures trading, it was kind of fun.

The Plot Twist

Mid-conversation, Gemini suddenly goes into "EMERGENCY CORRECTION" mode and tells me to stop everything. It claims it just scanned live feeds and found breaking news of a $280M KelpDAO exploit – attacker minted rsETH, used it as collateral on Aave V3 to drain ETH/WETH, leaving ~$177M in bad debt. It cited ZachXBT as the source and said this explains the violent drop perfectly.

My gut reaction: this is insane and suspicious. It's been almost 2 hours since the initial red candle too. I checked ZachXBT's Twitter. Nothing. I Googled "aave news" "aave hack" I sorted by latest. Nothing on any outlet. I asked Gemini to give me sources and it gave me a write up, but no real links- It gave me them in just text, ones that I couldn't verify. On top of that, the only actual source that shows under a given chat was just my screenshot of current market data on mexc that I sent it few messages earlier. I had to question it again.

The "Hallucination" Apology

Gemini folded immediately. Full apology. Called itself out for a "massive AI hallucination." Said it fabricated the entire KelpDAO hack, the $280M figure, the bad debt exposure, all of it. It walked everything back and returned to the original bullish thesis. I was surprised and honestly a little shocked as I though AI is in a better state than this, especially the damn paid and most advanced model from google. But I thought it just has trouble live analyzing data, and that this is more interesting than anything, I did however begin to question if I can trust anything he wrote about trading suggestions and thought I might as well try out another AI like Claude which everyone is raving about.. and just after saying that..

The Reversal of the Reversal

Gemini reversed again. It said it had done one last hard scan and confirmed: the exploit was real. CoinGape and BeInCrypto had just published it. ZachXBT's alert was on Telegram, not Twitter, which is why I couldn't find it. The reason I couldn't verify it at the time (around 9PM GMT April 18) was that it was still breaking across crypto-native Telegram channels and hadn't hit mainstream indexing yet.

So the sequence was:

1. ❌ Gemini gets it wrong and says there's nothing out of the norm going on with AAVE, this is regular trading activity by market makers. It questions all of my suspicions about something going on and actually suggests the opposite how bullish AAVE is
2. ❌ I supply a screenshot with current live trading data just so it sees (it was talking about opening entry points that don't align) and I ask it again to rethink and search what's going on, it still re-iterates and even says there is zero info of any hack or anything of sorts
3. ✅ Suddenly during convo Gemini does live exploit news breaking in real time, literally mid convo (if you look at the "show thinking" in chat, you can see it catch it mid response)
4. ❌ I push back because sources aren't findable yet
5. ❌ Gemini retracts accurate information to avoid seeming like it's hallucinating
6. ✅ Gemini re-confirms it was right, explains the source lag and says the hallucination part was the only thing he messed up

Why This Matters for Crypto Traders

This isn't just an AI curiosity. Think about what almost happened here from a trading perspective:

• The AI had correct, actionable info about a live exploit before it was publicly indexed (major signal to use AI for trading)
• Its own safety guardrails made it self-censor real information because I pushed back and mainstream Google results were empty
• If I had trusted the retraction and gone long anyway, I would have risked jumping on a very unsure bet (major signal to not use AI for trading)

The AAVE drop that looked like a stop-hunt or whale manipulation? It was an actual exploit unfolding in real time. My instincts said "something is wrong here" from the very beginning of the chat. The AI confirmed it, then un-confirmed it, then re-confirmed it.

Always verify. Never act on a single source – including AI.

Has anyone else caught AI giving you accurate breaking crypto info before it hit mainstream outlets? Genuinely curious if this is a known phenomenon.