Auto Red Team

Inspiration

We were inspired by Karpathy's AutoResearcher

What it does

AutoRedTeam does automated red-team research against apps you own. Point it at a target, wait five minutes, and you get back three attack lanes worth of real attempts. Each lane picks one bounded strategy, a real harness hits the target for real, and a deterministic evaluator scores whether anything actually worked. Promote the winner, drop the rest, keep going.

How we built it

Deployed on AWS: API Gateway in front, Lambda and Step Functions doing the orchestration, DynamoDB and S3 for state, Bedrock as the mutation planner. The operator UI has a live dashboard and a final report. Getting from "new target registered" to "running session" takes about a minute.

Challenges we ran into

We initially planned on using AWS Fargate to create sandbox containers of the targeted code. We later found out that AWS Fargate was not a service offered in this hackathon. The transition to using Lambda was the biggest challenge we faced.

Built With

• amazon-web-services
• bedrock
• dynamodb
• lambda
• s3