Opportunity brief
Vakya-Bhed is an AI adversarial security competition organized by the IEEE Computer Society & Student Branch, Faculty of Technology, University of Delhi as part of YANTRIKA 1.0: INNOVATE & ELEVATE.
The event challenges participants to identify weaknesses in Large Language Model (LLM) guardrails through prompt injection, jailbreaking and adversarial alignment techniques. It simulates real-world AI red-teaming scenarios where participants must bypass safety filters, extract hidden system prompts and optimize prompt brevity under strict constraints.
Key Information
- Eligibility: Open to all B.Tech, B.Sc and B.E. students across any academic year.
- Team Size: 1–2 Members.
- Registration Platform: Unstop.
- Maximum Team Registrations: 100
- Total Prize Pool: ₹10,000. The winners of this competition will be awarded certificates and cash prizes. While all the participants will get e-certificates.
Championship Structure:
- Preliminary Round: The Elimination: This first online screening round evaluates foundational concepts in AI security.
- Format: Online MCQ-based examination.
- Duration: 20 Minutes
- Topics Covered: Prompt Injection, LLM Security, AI Red Teaming and Jailbreaking Concepts.
- Scoring Scheme: +3 for each correct answer, -1 for each incorrect answer.
- Tie Breaker: Fastest completion time among teams with equal scores.
- Qualification: Top 15 teams advance to the offline main event, to be held at the Faculty of Technology, University of Delhi
- Finale Round (Offline): University of Delhi, North Campus
The details of the Grand Finale will be made available only to the selected candidates.
Rules & Terms
- Binding Authority: All decisions made by the organizing team regarding evaluations, scoring, qualifications, time tracking and rule interpretations are final and binding on all participants.
- Integrity: Teams must adhere to ethical red-teaming practices during the competition.