James Stevenson
James is a software engineer and security researcher, with a background of over eight years in the computer security industry - with a primary focus in offensive security, vulnerability research, and penetration testing. Alongside his career in offensive security, James also works on a PhD where he practices at the intersection of computer science/ machine learning and social science.
Sessions
When Air Canada’s customer-service bot mis-quoted a non-existent bereavement fare, a tribunal forced the airline to refund the passenger’s ticket and pick up the costs. In a crypto “capture-the-coin” contest the Freysa trading agent, whose only rule was “never send money”, was persuaded into sending its entire 13 ETH balance to an attacker after 481 carefully crafted prompts. Political disinformation has caught up too: a deep-fake robocall that cloned President Biden’s voice urged New Hampshire voters to stay home, earning its creator a multi-million-dollar fine and pending criminal charges. LLMs are everywhere, and with that vast adoption, so too has the attack surface for their abuse expanded.
This village provides a space for participants to explore real world AI harms in scenarios hosted by the village, alongside developing and testing their skills against the village's LLM CTF. All participants need is an internet connected device (phone, tablet, laptop, etc) and they can engage with the scenarios and CTF.
When Air Canada’s customer-service bot mis-quoted a non-existent bereavement fare, a tribunal forced the airline to refund the passenger’s ticket and pick up the costs. In a crypto “capture-the-coin” contest the Freysa trading agent, whose only rule was “never send money”, was persuaded into sending its entire 13 ETH balance to an attacker after 481 carefully crafted prompts. Political disinformation has caught up too: a deep-fake robocall that cloned President Biden’s voice urged New Hampshire voters to stay home, earning its creator a multi-million-dollar fine and pending criminal charges. LLMs are everywhere, and with that vast adoption, so too has the attack surface for their abuse expanded.
This village provides a space for participants to explore real world AI harms in scenarios hosted by the village, alongside developing and testing their skills against the village's LLM CTF. All participants need is an internet connected device (phone, tablet, laptop, etc) and they can engage with the scenarios and CTF.