AI investingportfolio reviewChatGPT promptsrisk managementred teaming

We Asked AI to Attack Our Own Portfolio. It Found the Weak Spot in Four Minutes.

Profile picture of Prateek Goel
Prateek Goel · INH00002639016 Jul 2026 · 3 min read · Updated 16 Jul 2026, 10:25 pm
We Asked AI to Attack Our Own Portfolio. It Found the Weak Spot in Four Minutes.

In our earlier AI article we made one claim that deserves a follow up: the single best use of AI in investing is making it argue against you. This week, with missiles in the headlines and crude near $85, we ran that exercise for real on one of our model portfolios. Here is exactly how it went.

THE SETUP

We gave the model the holdings of one ETH sector portfolio, each position's weight and our written thesis for it. Then one instruction: assume our thesis is wrong, and build the strongest case against every holding using only known, verifiable facts, ranked by how much damage each risk could do.

Four minutes later we had a ranked list of twelve attack points. Nine were risks we had already documented and sized for, which was reassuring rather than wasted effort, because a hedge you cannot articulate is a hedge you do not have. Two were generic filler of the markets can fall variety, which is the noise you always get from these exercises.

One was the reason this article exists.

WHAT THE MACHINE CAUGHT

The model pointed out that three of our holdings, in three different sectors, shared a single hidden dependency: all three needed crude oil to stay below roughly $90 for their margin assumptions to hold. Individually each position had sensible risk. Collectively, nearly a third of the portfolio was quietly short crude without us ever writing that sentence down. In a week where Brent is climbing on Gulf tension, that is not an academic observation. We resized nothing dramatic, but one position's weight came down at the next rebalance, and the crude dependency now sits in the risk notes where it always should have been.

No stock pick. No prediction. Just a correlation we were too close to see, surfaced by a machine that reads faster than we do and owes our thesis nothing.

RUN IT ON YOUR OWN PORTFOLIO

The prompt structure that works: list your holdings with weights. State your reason for each in one line. Then instruct: act as a sceptical risk analyst, assume my reasoning is flawed, attack each holding using verifiable facts, rank the attacks by potential damage, and finally identify any hidden dependency shared across multiple holdings. That last clause is where the value lives. Any of the large AI chat tools can run this today, free.

THREE HONEST WARNINGS

Verify every fact before acting, because models still confidently invent numbers. Never paste anything you would not want leaked; holdings and weights are enough, account details are never needed. And treat the output as a checklist for your judgment, not a replacement for it. The machine found our crude exposure. Deciding what to do about it remained a human job, done with written reasons, the way every decision at this desk gets made.

AI will not pick your stocks. But once a quarter, let it try to break your portfolio. It is the cheapest risk audit available in India today.

You own the stocks. We write the reasons. Every single time.

DISCLAIMER: This article is for education and information only. It is not investment advice, not a recommendation, and not a solicitation to buy or sell any security. The exercise described is illustrative of process, not a performance claim. Prateek Goel is a SEBI Registered Research Analyst (INH000025902). Investments in the securities market are subject to market risks. Read all related documents carefully before investing. Registration granted by SEBI and certification from NISM in no way guarantee the performance of the intermediary or provide any assurance of returns to investors.