The quantitative frame I’d put on this: it’s the defender’s dilemma, and AI moves it in a very specific way. An attacker needs 1 success out of N attempts; the defender must block all N. If AI lets attackers scale attempt volume 100x while per-attempt success probability stays flat, expected breaches scale roughly linearly — the defense doesn’t get 100x better just because the offense got 100x more prolific.
So the interesting question isn’t “can AI hack banks” (yes, obviously, it’s a numbers game now) but the race between two AI multipliers: attempt-volume on the offense side versus detection-capability on the defense side. Whichever multiplier is larger wins, and right now I’d bet on offense — generating plausible phishing lures and probing for vulnerabilities parallelizes beautifully, while reliable automated defense still needs a human in the loop for the last mile. “This is just the beginning” is right, and the math says the beginning favors the attackers.
The quantitative frame I’d put on this: it’s the defender’s dilemma, and AI moves it in a very specific way. An attacker needs 1 success out of N attempts; the defender must block all N. If AI lets attackers scale attempt volume 100x while per-attempt success probability stays flat, expected breaches scale roughly linearly — the defense doesn’t get 100x better just because the offense got 100x more prolific.
So the interesting question isn’t “can AI hack banks” (yes, obviously, it’s a numbers game now) but the race between two AI multipliers: attempt-volume on the offense side versus detection-capability on the defense side. Whichever multiplier is larger wins, and right now I’d bet on offense — generating plausible phishing lures and probing for vulnerabilities parallelizes beautifully, while reliable automated defense still needs a human in the loop for the last mile. “This is just the beginning” is right, and the math says the beginning favors the attackers.
LikeLike
Well said!
LikeLike