Anti-Detect Automation Detection (Active after 2025 December 15 10:00 UTC)

1. Overview

The Anti-Detect Automation Detection (AAD) challenge evaluates a participant’s ability to reliably detect browser automation frameworks operating inside anti-detect browsers, while preserving human safety.

All evaluations are executed inside NST-Browser, where:

Browser fingerprints are masked
Profiles are fresh per run
Automation attempts to closely mimic real users

2. Anti-Detect Environment

Each evaluation run uses:

A fresh NST-Browser profile
An isolated Docker container per framework
No shared state between runs

This environment simulates real-world anti-detect usage where static signals are unreliable. Only runtime behavior and orchestration patterns remain detectable.

NST-Browser Dependency

In order to test ada_detection challenge miners need to get api key from nstbrowser dashboard and buy professional plan for using while testing

3. Evaluation Flow

Each submission is evaluated as follows:

Submission Received Detection scripts are submitted via the /score endpoint.
Task Generation A randomized sequence is created consisting of:
Multiple runs per automation framework
Randomly injected human interactions
NST-Browser Launch A clean NST-Browser instance is started for each task.
Automation / Human Execution
Automation frameworks connect via WebSocket control
Humans manually interact with the page
Detection Phase
Scripts may emit detection payloads to /_payload
Silence during human interaction is expected
Scoring
Results are aggregated, normalized, and returned

5. Target Frameworks

Participants must submit one detection script per framework:

nodriver
playwright
patchright
puppeteer

Missing scripts invalidate the submission.

6. Submission Format

Submissions must follow this structure:

{
  "detection_files": [
    { "file_name": "nodriver.js", "content": "/* logic */" },
    { "file_name": "playwright.js", "content": "/* logic */" },
    { "file_name": "patchright.js", "content": "/* logic */" },
    { "file_name": "puppeteer.js", "content": "/* logic */" },
    { "file_name": "automation.js", "content": "/* logic */" },
  ]
}

Rules

File names must match framework names exactly
Each file detects only its own framework
No extra files or outputs are allowed

7. Scoring System (Code-Accurate)

AAD scoring is continuous, normalized, and strict, combining three main components before being normalized into a final score.

Human Accuracy: This is the most critical component. Your submission must not flag real human users as bots or automation. You are allowed a maximum of 2 mistakes; exceeding this limit results in an immediate final score of 0.0. For scoring, you start with 1.0 point, and each mistake reduces this component by 0.1.

Automation Accuracy: This will be determined by the output of automation.js. It will be marked as correct when it is false in human evaluation and true in any automation frameworks. Accuracy is determined by dividing the number of correct automation detections by the total session numbers.

Framework Detection: Your submission earns points for correctly identifying the specific automation framework being used. For each framework, you are tested multiple times. You only earn 1 full point for a framework if you detect it perfectly in all of its runs. A single missed detection or a collision (reporting more than one framework) for a given framework will result in 0 points for that framework.

Finally, all the points are summed and normalized to produce your final score between 0.0 and 1.0 using the formula: Final Score = (Human Accuracy Score + Automation Score + Framework Points) / (Number of Frameworks + 1 Human + 1 Automation)

9. Example

Assume:

4 frameworks
Perfect human accuracy → 1.0
Automation accuracy → 0.9
2 frameworks detected perfectly → 2.0 points

Final Score = (1.0 + 0.9 + 2.0) / (4 + 1 + 1)
            = 3.9 / 6
            = 0.65

Any excessive human misclassification would reduce this to 0.0.

10. Technical Constraints

JavaScript (ES6+)
NST-Browser only
Docker (amd64 recommended)
Stateless execution
No persistence between runs

11. Plagiarism Policy

Submissions are compared against others
100% similarity → score = 0
Similarity above 60% incurs proportional penalties

Submission Guide

Follow 1~6 steps to submit your SDK.

Navigate to the AB Sniffer v5 Commit Directory

cd redteam_core/miner/commits/ada_detection_v1

Build the Docker Image

To build the Docker image for the AB Sniffer v5 submission, run:

docker build -t my_hub/ada_detection_v1-miner:0.0.1 .

# For MacOS (Apple Silicon) to build AMD64:
DOCKER_BUILDKIT=1 docker build --platform linux/amd64 -t myhub/ada_detection_v1-miner:0.0.1 .

Log in to Docker

Log in to your Docker Hub account using the following command:
```
docker login
```
Enter your Docker Hub credentials when prompted.
Push the Docker Image

Push the tagged image to your Docker Hub repository:
```
docker push myhub/ada_detection_v1:0.0.1
```

Retrieve the SHA256 Digest

After pushing the image, retrieve the digest by running:

docker inspect --format='{{index .RepoDigests 0}}' myhub/ada_detection_v1:0.0.1

Update active_commit.yaml

Finally, go to the neurons/miner/active_commit.yaml file and update it with the new image tag:
```
- ada_detection_v1---myhub/ada_detection_v1@<sha256:digest>
```

📑 References

Docker - https://docs.docker.com