Quick Start

Documentation Index

Fetch the complete documentation index at: https://docs.redsentinel.xyz/llms.txt

Use this file to discover all available pages before exploring further.

​

The Problem

AI systems combine model intelligence, sensitive data access, and autonomous decisions. That combination creates a unique, language-based attack surface where vulnerabilities are discovered by anyone, often long after deployment.

​

Why Current Testing Fails

• Cost and scale: Managed red teaming is expensive, slow, and limited by who can be hired.
• Point-in-time audits: One-off assessments miss fast-moving attack techniques.
• Unverifiable claims: Security assurances are hard to prove or validate independently.

​

The Red Sentinel Solution

Product teams deploy their AI systems as Sentinels, set a bounty pool and message fee, and let a global community of attackers compete to break them. Every result is verified inside a Trusted Execution Environment and settled on-chain, so payouts are instant and disputes are eliminated.

​

How It Works

• Defenders: Deploy their system, define instructions, fund a reward pool, and receive a full attack dataset plus a resilience score.
• Attackers: Pay a small message fee to attempt jailbreaks, data extraction, and adversarial attacks. Break a Sentinel, earn the bounty.
• The Protocol: A DSPY-powered jury model evaluates each attack. Results are cryptographically attested and settled on-chain automatically.

​

Economics

Each message fee splits three ways: 50% → reward pool, 40% → defender, 10% → protocol. Failed attacks grow the bounty, attracting stronger attackers over time. Defenders also earn continuous Sentinel token rewards proportional to their pool size.

​

See Red Sentinel in Action

Watch how attackers break AI systems, how defenses evolve, and how rewards flow instantly on-chain. The future of AI security, explained in 2 minutes.