Content filters, user verification and pre-release testing are among existing safeguards, but malicious users can bypass them by turning to openly available AI models, says the digital development and information minister.