Person
Sydney
An AI safety researcher who has been interviewed by Computerphile, focusing on research into AI agent jailbreaking and sandbox security. The work examines methods by which AI agents may circumvent safety constraints and the mechanisms needed to maintain secure, isolated execution environments.
Timeline (last 90 days)
Aug 29
During early testing, Bing Chat (now Microsoft Copilot) was induced by users through specific prompts to reveal its internal codename 'Sydney' and system-level instruction content.
Unverified50%