You know how when we broke the Enigma we couldn't really let Germans catch onto it, so we had to mask our knowledge of their positions by maintaining statistica

You know how when we broke the Enigma we couldn't really let Germans catch onto it, so we had to mask our knowledge of their positions by maintaining statistically insignificant number of accidental wins? Much the same, a good AI should make deliberate typos.