Not me, but a colleague at a former job made polyaniline by a method involving the use of a perchlorate salt. Having filtered off the polymer, he noticed that the filtrate was coloured, and assuming some of the polymer was still in solution, and being unwilling to lose any, he evaporated off the water in a dish and added the residue to the filtration residue, and put this in a vacuum oven to dry overnight. It blew up the oven (fortunately at night when nobody was in the lab).
At university, someone (debate still rages over who) managed to put a large steel adjustable spanner into a bowl of aqua regia. When we returned after coffee, it was belching out NO2 like a steam train, fortunately it all went up the fume hood. We were more concerned about the loss of the expensive spanner than the chemical hazard!
On the subject of NO2, I once had to make LiCoO2 (by a way that was supposed to give better properties than the commercial material). Started with reagents in solution in a 25L reactor; as the water evaporated and reaction proceeded, the mixture got thicker, and as it reached the gel point it bubbled up (NO2 was evolved) and pink foam (full of carcinogenic Co) overflowed uncontrollably from the reactor like something out of the Sorcerer's Apprentice. Though wearing a lab coat, the right cuff of my blue shirt was bleached white by acidic gases (from going into the reactor with a long spoon to check the state of the mixture). I wanted to abandon the proedure but boss insisted on proceeding.
An interesting little curiosity: I once made an azobenzene-terminated polymer (polysulfone of some kind I think) via diazotisation of the amine-terminated polymer. I used it to test some GPC systems we were thinking of buying. The idea was that at 254 nm the signal would be proportional to the mass concentration (being due to the benzene rings in the backbone), while at the azo wavelength (I forget, perhaps about 400 nm) it would be proportional to the molar concentration (as each molecule has just 2 azo groups). So the chromatograms ought to give the weight fraction and number fraction distributions respectively. Worked quite well, if I remember, though not perfectly.