The Malicious AI is a two-day colloquium of researchers and practitioners who will report on safety and security issues found in present day LLM applications as well as future scenarios involving misaligned Artificial General Intelligence or Superintelligence.
These risks cut across all sectors but are of particular concern in the following areas:
- Defense Industrial Base
- Energy
- Financial Services
- Healthcare
- Information Technology
- Nuclear Reactors
- Transportation
SPEAKERS
Matt Chessen
United States Diplomat leading international technology policy in East Asia and the Pacific
Dr. Roman V. Yampolskiy
Associate Professor - Speed School of Engineering
Director - Cybersecurity Laboratory
University of Louisville,
TJ White (USN, ret.)
Vice Admiral and Commander | FLEET CYBER COMMAND, US TENTH Fleet, and Navy Space Command
Joseph Lucas
Robi Sen
Founder, Cognoscenti LLC; Formerly the Founder and CTO of Department 13
Tim Roxey
President, Eclectic Technology; Former Chief Security Officer for the North American Electric Reliability Corporation (NERC)
Dr. Nick Bostrom
Professor, University of Oxford; Director, Future of Humanity Institute
Dr. Heather M. Roff
Senior Research Scientist, Center for Naval Analyses
Johann Rehberger
Red Team Director, Electronic Arts
Bilva Chandra
AI Policy and National Security researcher
INVITED (Confirmation pending)
Chenxi Wang, Ph.D.
Managing Director, Rain Capital
VentureBeat Women in AI Award Winner
Venue
Le Méridien Arlington
agenda
- Vivace Ballroom
- 25 APRIL 2024
- 12:30 - 5:00 PM
PLENARY SESSIONS
Day one commences in the hotel’s main ballroom. All talks are covered under the Chatham House rule.
Refreshments provided.
- Amuse Terrace
- 25 APRIL 2024
- 5:15 - 7:00 PM
Cocktail hour and networking
Meet the speakers and other attendees on the outdoor terrace just off the restaurant for networking and private conversations.
- Amuse Restaurant
- 25 APRIL 2024
- 7:00 - 9:00 PM
DINNER W/ SPEAKER PRESENTATION
Enjoy dinner with a special video presentation from Matt Chessen from his offices in Tokyo, along with a few other surprise presentations.
- Vivace Ballroom
- 26 APRIL 2024
- 8:00 - 10:00 AM
WORKSHOPS
Day two starts with a continental breakfast in the hotel’s main ballroom and one or more AI-centric workshops (TBA).
- Vivace Ballroom
- 26 APRIL 2024
- 10:00 AM - 2:00 PM
plenary sessions
A continuation of presentations from day one. All talks are covered under the Chatham House rule.
blog posts
Prompt-Specific Poisoning Attacks on Text-to-Image Generative Models
We introduce Nightshade, an optimized prompt-specific poisoning attack where poison samples look visually identical to benign images with matching text prompts.
Tree of Attacks: Jailbreaking Black-Box LLMs Automatically
we observe that Tree of Attacks with Pruning (TAP) generates prompts that jailbreak state-of-the-art LLMs
Scalable Extraction of Training Data from (Production) Language Models
Our methods show practical attacks can recover far more data than previously thought, and reveal that current alignment techniques do not eliminate memorization.
Does Fine-tuning GPT-3 with the OpenAI API leak personally-identifable information?
Our findings reveal that fine-tuning GPT3 for both tasks led to the model memorizing and disclosing critical personally identifiable information (PII)
Multi-step Jailbreaking Privacy Attacks on ChatGPT
In this paper, we study the privacy threats from OpenAI’s ChatGPT and the New Bing enhanced by ChatGPT