David Robinson, who recently resigned from OpenAI after 3.5 years, wrote in the Atlantic that the company and its competitors are not "nearly careful enough" with AI safety. Robinson helped draft OpenAI's preparedness framework and reviewed safety reports for 12 major model launches before leaving.
Robinson's main concern: OpenAI uses "iterative deployment," releasing AI systems and then fixing safety problems as they emerge, rather than thoroughly testing safeguards beforehand. He argues this approach is no longer appropriate as AI systems become more powerful. OpenAI responded by stating it pauses or delays model launches when safety concerns arise.
Why this matters for Singapore and Hong Kong families
- AI tools like ChatGPT are already used by students and schools; understanding their safety gaps helps parents assess risks when children use them.
- Safety failures in commercial AI systems could affect data privacy and reliability of AI tools schools adopt for learning.
- The debate signals that AI regulation and industry safety standards remain unsettled, affecting what guardrails eventually protect users.
What to watch
- Check whether schools using AI tools have clear policies on data handling and model limitations before signing consent forms.
- Review any AI safety incident reports OpenAI or other providers publish to understand what problems have actually occurred.