No idea, I'm not a lawyer. Assuming liability is a real problem presumably this is fixed by ad networks improving their vetting and offering some sort of insurance to publishers in the event malware sneaks through.

Alternatively, as mentioned in another comment, proxying might mean simply installing a managed ad-server within the publisher's infrastructure. So from an end-user PoV it looks like it's coming from the publisher's server but in reality it's still being served from a third party managed box.

Until the adserver serves malicious ads (which has happened dozens of times on high profile sites in the past year or so), and then the website is sued.

Yeah but if the adserver is a black box that's managed by the ad network I would have thought the website could pass the liability on to them (and state this in their ToS somewhere)