This touches on the much more likely scenario whereby the car's internal data doesn't mesh with the data from it's sensors. If the sign says "one way" it is a one way street. What google maps says doesn't matter. Current warning signs trump previously recorded data. "But google maps said I could do 70!" will be no defense.
There is no doubt that intelligent highways would be a massive target for terrorist and cyber attacks. However, one thing in our favour is the fact there is so little technology already in place, since they won't be burdened with decades' worth of insecure legacy systems: at least they'll be able to start out building infrastructure in full knowledge that it better be secure.
You have your own trusted sources of information, and you have your secondary sources of information, which you apply trust levels to. Your own trusted sources of information are your own senses, your sight, hearing, feeling (touch), smell, etc. You may have something communicated to you about your condition or environment through these senses, but it's silly to take that at face value without confirming with your own senses. If your friend next to you that you trust tells you that you're about to walk into a wall, you look ahead. If you can't see a wall, you might slow and exercise caution until you figure out why you were relayed that information, but unless you can completely stop without problem, you don't do that without cause.
I think it's a mistake to think of cards in the future communicating as a swarm. They need to be able to function independently, and also to take in extra information from the group when the group is available, and make decisions on that. That's less swarm behavior than social behavior, so we should consider groups of cars on the road a social groups, and the same information dynamics exhibited in those groups apply.
1: http://gizmodo.com/how-a-hacker-could-hijack-an-airplane-fro...
2: https://www.wired.com/2015/07/hackers-remotely-kill-jeep-hig...
US voting machines didn't have to deal with legacy technology and still turned pretty bad from an infosec POV.
The lesson we do need to learn is future-proofing: whatever infrastructure we install on the millions of miles of public highway will need to be serviceable for decades.
http://www.nhtsa.gov/staticfiles/rulemaking/pdf/V2V/Readines... (PDF link)
Say your car A trusts other cars B, C ... Y to varying degree in a WoT. It meets a brand new car Z that is not yet trusted by any cars A-Y. Car A receives a message from Z, to which it assigns 0 credence points, meaning this data is not taken into account in any decisions. If car A then makes an observation confirming the message, car A now trusts Z by +1 point, and lets other cars know this. (I'm unsure how to make aggregation/distribution of trust work smoothly, but I believe the random mixing of traffic patterns will work in the system's favor.)
So gradually car Z becomes trusted, up to some maximum, e.g. 100. The only worry then is a malicious actor who mostly sends true messages, but that's hard to distinguish from a car with a sometimes faulty sensor. Those problems could be handled the same way, namely aggressively penalising a car for sending incorrect data (e.g. by -30), such that it's quickly distrusted, together with a prominent warning in the failing car when it goes below 0 trust that it needs maintenance (after which trust would be reset to 0). Possibly also increased insurance premium if your car remains with a trust below 0.
This sounds very much like you're describing an antivirus program, which we all know works perfectly well /s