OpenAI’s latest GPT-5.6 safety results show low failure rates for direct prompt injection but higher success rates when attacks arrive through tools and external content. If you can only read one tech ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results