Researchers call for stricter AI standards and testing as harmful responses like hate speech and copyright violations surge