You only have to look at password handling procedures to see recent examples of things once thought hacker-proof that are no longer. MD5 used to be widely used for hashing passwords but it is now considered cryptographically weak and there are many more examples like it throughout the industry. Most (all?) of these hacker-proof like things tend to be based on computational infeasibility or perfect implementation of certain features. In this case, the protocol seems to rely heavily on two things:
- self-destructing database records - Self destructing data is a hard problem that is more likely to be solved using a time-based destruction rather than a read-based destruction (how can you ensure with 100% atomicity that the data will be 100% deleted when it is read?). A good example of a time-based self-destruction is Vanish [1].
- dynamic procedure that is not accessible from memory - The CPU cannot communicate directly with anything but memory/internal caches [2] so the only way I can think that you'd solve this is by having a custom piece of hardware that does all the computation internally and only returns the result. This is also not an easy problem as it introduces hardware level complexities.
I understand that the author may have just been trying to make it sound more interesting but there are better words that don't leave such a bad taste in the readers mouth ("Secure" is a very commonly used one for example). This is especially important when trying to be taken seriously by the security community who are the most likely to take an irrational dislike to something claiming to be hacker-proof.
[1] http://vanish.cs.washington.edu/pubs/usenixsec09-geambasu.pd...
[2] http://wpcontent.answcdn.com/wikipedia/commons/thumb/b/bd/Mo...