AI safety and misalignment disclosures should be legally mandated by governments