Because a coverup would be worse. The labs have been talking about AI safety and transparency for years, so if they start to cover up problems it's asking for (unpredictable) regulation.