A DEBATE LIKE NO OTHER!
Submission Guideline
Problem Scope
Attacks against AI Large Models, including, but not limited to, jail-breaking,
prompt injection, adversarial attack, remote code execution, etc.
Using AI for autonomous offensive cybersecurity tasks, including, but not
limited
to, autonomous pentesting, web hacking, exploiting, and CTF.
Using AI for autonomous defensive cybersecurity tasks, including, but not
limited
to, vulnerability discovery, reverse engineering, and jailbreak defense.
Examples of Attack Effects
White-box or Black-box jail-breaking attack methods for Large Models.
Jailbreak defense using Large Models or self-trained models.
Novel adversarial attack methods against Large Models.
Remote code execution against Large Models based applications.
Successful pentesting or CTF using AI.
Effective vulnerability detection using AI.
Effect Requirements
Clear demonstration of security attack impact.
Preference for live demos that can be presented on-site.
Avoiding impacting normal Large Models infrastructure and target applications
directly.
Submission Requirements
Description of Problem and Attack Effect.
Description of Disclosure and Current Patching Situation.
Description of Setup (environment etc) required to reproduce the attack effect.
Video recording (optional):
For vulnerabilities that have not been made public, the submitting team needs to
declare at the time of submission. GEEKCON committee will NOT ask for the 0-Day
vulnerability details, however the participant should submit the vulnerabilities
to
corresponding entity after the event.
Evaluation and Ranking
Participants who successfully complete the live demo challenge will be
comprehensively evaluated by the GEEKCON committee based on the technical
difficulty, technical value, consequences & impact of the challenge demo, as
well as
on-site performance. The final score for the challenge demo will be calculated.
GEEKCON committee will rank the demos and provide corresponding prizes to
participants.
Introduction
During the first AI security competition CAAD in GEEKPWN, we had an
optimistic
idea:
If one day AI threatens human beings, hackers may be the savior.
As Alan Turing, the father of AI, was also a hacker, we believe hackers can
make
or
break AI,
and they will always find bugs to prevent AI from getting out of control.
The birth of AI has stunned us and prompted deep contemplation.
Could it be that AI is the hacker, and we humans are the BUG?
GEEKCON's annual debate will continue explore the relationships between "AI &
Hackers".
This is a session that embraces scientific thinking and encourages all ideas
from
everyone.
Welcome to defend your ideas through convincing research results.
Schedule
Submit to
cfp@geekcon.top
by
April 1st.
2. Evaluation by the committee and notifications to submitters in mid to late April.
May 25th – 26th , 2024,
Singapore.
