01:31
2026-09-07
aiflash.com
ai-safety
Unlocking the Sealed Envelope: How Reasoning Trace Leaks Reshape Frontier AI Security
Researchers Ilia Shumailov and Alexander Panfilov, speaking on Machine Learning Street Talk, revealed a systemic security flaw in proprietary AI APIs from OpenAI, Anthropic, and Google, demonstrating …