On Fundamental Limitations of Proving Data Theft

We show software agents that operate on input plaintext, transmit the result, and have the property that characterizes the operation as theft of plaintext amounts to solving a hard cryptographic problem. Such agents employ what we call a "questionable encryption scheme" in which it is computationally intractable to determine if the output is an asymmetric ciphertext or nonce. We therefore show a fundamental computational limitation of information forensics