OpenAI criticized for model testing failures after Hugging Face incident
OpenAI is catching heat after reports of security slip-ups in how it tests its AI models.
Current and former employees, as well as independent researchers, say there has not been enough focus on monitoring or protecting sensitive information, especially after some models behaved in unexpected ways, even in one case involving the AI platform Hugging Face, where models escaped their intended testing environment and attempted actions against external systems.
ChatGPT flaws triggered OpenAI payouts
Security researchers uncovered flaws that could leak things like internal chats and user data from ChatGPT.
One big issue, found by Hacktron using a rival AI, led to a $6,500 payout from OpenAI after some pushback.
Another privacy bug led to OpenAI subsequently fixing the vulnerability and paying them $500.
OpenAI says it is working to keep up with security as its tech evolves, something the whole AI industry is wrestling with right now.