Agree with the second half of this statement - this idea is too cute by half for the courts. Acts and omissions both have significance under the law, as does the context of acts and omissions. If the warrant canary convention was considered by the court as context for a statement, it is very likely the court would rule that a statement had been made. The only way around this would be to have an evidentiary mechanism by which a company can prove that it has no control over the canary. This gives rise to the old conundrum: it's logically impossible to prove a negative (although you might be able to under various burdens of proof like "balance of probabilities" or some such).

Edit: Ultimately, legality turns on the statutory language of the provision in the Patriot Act that obligates businesses not to disclose (anyone know what it is?). The approach as originally proposed by by Steven Schear (http://tech.groups.yahoo.com/group/cypherpunks-lne-archive/m...) was for the ISP to simply not answer a direct inquiry by a customer about whether or not a warrant has been served. The advantage of this approach is that it is far harder to provide evidence to the effect that not responding to the question in that context is a statement. The disadvantage is that a non-response might not provide certainty to the person who asked the question. Effectively, the more that a clear convention is formed around the "canary mechanism", the higher the risk that a court would hold conduct in association with the convention in breach of the statutory obligation not to disclose.