Jonathan Rodriguez Cefalu

Helping save the lives of elderly patients by providing early detection of life-threatening respiratory complications, heart failure, and infections.

Developing solutions to ensure human control of increasingly powerful artificial intelligence (AI) systems. I was first in the world to discover/execute any prompt injection attack against any LLM, and I have the first patent on prompt injection defense. See https://research.nccgroup.com/2022/12/05/exploring-prompt-injection-attacks/

I led this high-profile hack and designed the prompt injection payload for it: https://www.theregister.com/2024/08/21/slack_ai_prompt_injection/ I also developed two even more serious attacks against bigger enterprise targets that remain vulnerable and as such I have not published yet.

Much of this work is unpublished for military reasons, but some is declassified here: https://simc4.org I am the ML Research Lead for this initiative: At DISARM:SIMC4, (an all-volunteer Track 2 working group) we attempt to map and understand the risks of escalatory feedback loops in an increasingly digitized military landscape. Our core function is to identify the new risks associated with the involvement of A.I. in decision support (such as automated threat detection), especially risks that could result in a devastating "flash war", an analogy with the 2010 stock market "flash crash" wherein a "flash war" is an accidental war or accidental escalation of a conventional war beyond the nuclear threshold. Additionally, we promote policy measures such as the Decide Under Attack nuclear posture concept which will reduce the vulnerability of the great powers to being tricked into fighting a nuclear war by terrorists using a false flag. Such terrorists would create false nuclear missile warnings (via cyberattacks, A.I. data spoofing, deepfakes, and compromised personnel) to try to trick the great powers into destroying each other - which might be perceived by some radicals as a clean slate allowing the creation of a new state or caliphate. Finally, we seek to serve as a forum where military A.I. researchers from rival powers can conduct open dialogues to propose ideas for stabilizing technical measures (such as methods of reducing the vulnerability of A.I. to spoofing) which might reduce the risk of accidents and reduce the vulnerability to false flag terrorism.

Leading system architecture R&D across a truly full stack from display optics to graphics software to computer vision software to camera optics. I won Forbes 30 Under 30 for this phase of my career. https://www.forbes.com/profile/jonathan-rodriguez/ "Rodriguez was a cofounder at Vergence Labs, which was acquired by Snapchat in 2014 and became the foundation for its Spectacles project. Now launched, Spectacles, which are video-capturing glasses, is the social media company's first foray into hardware. As the architect of that project, Rodriguez, who dreamed up the device for his college honors thesis, is responsible for its design and function." -Forbes

Acting CTO + R&D lead for augmented reality pioneer Vergence Labs, Snapchat's first ever acquisition. We became the Snap Lab team within Snapchat, inventing Spectacles augmented reality eyewear! https://spectacles.com