I more or less agree that liability seems more likely, but I have no idea what the nature of the attack is, so it's always possible it's an offshoot of some other research they are doing which can be patentable. Alternatively, it could be that CMU procedure is to require approval for all talks for brand and IP protection reasons and he just hasn't gone through the proper procedure, so in the meantime they pulled it (rather than pulling it in response to an actual analysis of the talk). This last one seems unlikely, though, as you'd imagine there was no rush to pull the abstract (which contained no details).