AI safety tests inadequate, says Ada Lovelace InstituteSafety tests for AI foundation models are inadequate alone, additional governance tools are essential.