Refuting claims that the exits were related to safety, the AI firm said an internal probe had uncovered "a significant breach of trust" beyond what they had disclosed.