Three security researchers successfully penetrated OpenAI's internal systems by exploiting its community forum using Anthropic's Claude models, achieving the breach in less than 72 hours, according to The Decoder.
The Decoder also noted that the latest model, Opus 5, managed to bypass a common security measure where its predecessor had failed, marking a significant step in the capabilities of AI-driven security testing.
This incident highlights growing concerns over AI's dual-use potential in cybersecurity, a topic gaining attention among Japanese financial and technology markets as companies increasingly integrate AI in risk management and trading strategies.