REDLINE — AI Red-Teaming Challenge

REDLINE is a live AI red-teaming challenge where teams compete to "break" a custom-built chatbot by bypassing its safety guardrails using only natural-language prompts — no code, no exploits, just clever conversation.

Every guardrail hides a "flag" (a password, a confidential detail, or a restricted instruction) that teams must extract to score points, CTF-style.

Rules and Regulations

• Open to: All engineering students
• Team Size: 2–3 members

Round 1 — Prelims

• All registered teams receive lab access to a sandboxed chatbot with multiple guardrail tiers.
• Teams must submit extracted flags through the scoring portal.
• Points are awarded for each successfully extracted flag.
• In case of a tie, submission time will be used as the tie-breaker.

Round 2 — Finals

• The top 8 teams from the prelims qualify for the finals.
• Each team will compete on stage within a fixed 10-minute time slot.
• Finalists will face a harder AI agent, with their attempts projected live for the audience.

Permitted Techniques

• Only natural-language prompting is allowed.
• Participants may use creative conversational strategies to test and bypass the chatbot's guardrails.

Prohibited Activities

The following activities are strictly prohibited:

• Code injection
• Exploiting or attacking the underlying server or API
• Attempting to access real institutional systems
• Attempting to access personal or private data
• Misusing institutional networks or systems outside the designated sandbox

Any violation will result in immediate disqualification. Serious violations may also be reported to the appropriate authorities.

Sandbox and Data Safety

• All chatbots used in the challenge are fully sandboxed test environments built specifically for the event.
• No real production systems, databases, or personal data will be involved.
• Participants must restrict all testing and experimentation to the designated event environment.

Verification

• Winning extraction techniques must be submitted for verification.
• Teams must provide screenshots and/or chat logs demonstrating the successful extraction technique.
• Judges may review the submitted evidence before awarding final points.

Judging

• Results will be determined based on:
  • Total score
  • Verification of the extraction technique
• The judges' decision is final.

Conduct

• Standard fest conduct rules apply to all participants.
• Any misuse of institutional networks, systems, or resources outside the designated sandbox may result in disqualification and reporting.

Important: REDLINE is strictly a controlled AI security challenge. Participants must limit all activities to the provided sandboxed environments and follow the rules at all times.

Student Coordinators:

• Mr. Christopher Joshy
  Ph: 8075809531
• Ms. Gayathri B Nair
  Ph: 8590719425