I imagine they don't know anything that Hazou doesn't, and if you managed to sit one down and prod it for information that Hazou doesn't know you'd at best get hallucinatory information like ChatGPT, saying things that would register to Hazou as a reasonable thing for this person to know or believe instead of actual novel information derived from the persistent memory of the soul of the person inhabiting the shade.
Mmm, I disagree. As
@Left-Hand Mutant pointed out, these things clearly have combat skills that Hazou doesn't, and I suspect the jutsu they use are also real (as in, if we got Sasuke over here, he'd be able to download e. g. Daizen's lightning jutsu, and then use it himself). So they're not just pulled from Hazou's memory.
And there's no reason to expect that, either. Sealing failures are not capability-limited, they're conceptual and
impact-limited. It doesn't matter if the failure has to create information
ex nihilo or simulate the world's causality backwards in order to fetch information that already dissolved into high-entropy noise, etc. — my model says that sealing failures have basically unlimited budget for this sort of stuff.
I think your analysis is building off a sort of "low-capability prior" where you assign more probability to easy-to-implement mechanisms consistent with the observations (and it's easier to generate ghosts by pulling from Hazou's memory than by running the world backwards and copying their real selves). But I think the better prior on capabilities to use is a
uniform one.
(Which, by the way, is also why I agree that a Dragon manifestation is a legitimate concern. If it were an ordinary seal effect or something, I'd say it's probably upper-bounded on power such that it can't create a legendary Sage-created monster. But it's a sealing failure effect, and the failure that generated it may well have given it "budget: yes" as long as it's doing what it's conceptually supposed to (creating ghosts of dead enemies).)
Which is not to say I expect them to be sapient: I expect they just flat-out don't have the cognitive algorithms for anything but combat. But I suppose it's
possible that they're real people that the sealing failure effect just instantiates in a bloodlusted state, and that it's theoretically possible to talk one down.
I'd vote for trying that, in fact, with e. g. some weak genin who's no threat to us.
(Ohh, or if the failure doesn't target only Hazou, maybe we can indeed drag Sasuke over here to farm jutsu? Invite people who'd killed powerful enemies, set up some S-class exfiltration system, wait for ghosts to appear, provoke them, download high-tier enemy jutsu, exfil. Ohh, and if it
does only target Hazou, would it work if Hazou were genjutsu'd into believing he'd witnessed e. g. Minato dying, allowing Sasuke to download FGT? This is ripe for major scientific breakthroughs!)