⚠️ Responsible Disclosure: All vulnerabilities described here were reported through authorised channels — OpenAI's bug bounty programme on HackerOne — or are publicly disclosed findings from credited researchers. Never test production AI systems without written authorisation. OpenAI's Terms of Service explicitly prohibit unauthorised security testing of their API and products. ChatGPT has 200 million weekly active users. Every one of them is interacting with a system that, until researchers started testing it seriously, had never been through a rigorous adversarial…
No comments:
Post a Comment