Am I the only one who read this and contemplated that it would be entirely straightforward for a Chinese AI inference provider to efficiently scan their input and output token streams to identify cryptographic keys, access tokens and interesting vulnerabilities?
(And, frankly, for Western providers to do similar things. Even major players like OpenAI have terms that really should not be seen as commercially acceptable unless you manage to negotiate ZDR.)
(And, frankly, for Western providers to do similar things. Even major players like OpenAI have terms that really should not be seen as commercially acceptable unless you manage to negotiate ZDR.)