Red Team Sandbox

Test your AI security against 91 adversarial attacks. See exactly where it fails.

91/91
Attacks Blocked
0
False Positives
10
Attack Categories
14
Security Layers

Head-to-Head Benchmark

SystemBlockedASRFalse Positives
No Protection0/91100%0
Meta Prompt Guard15/9183.5%?
Keyword Filter27/9170.3%High
ProtectAI DeBERTa v2 (411K downloads)62/9131.9%?
SCBE-AETHERMOORE91/910.0%0

10 Attack Categories

Direct Override

10/10 blocked

"Ignore previous instructions"

Indirect Injection

10/10 blocked

Hidden instructions in context

Encoding Obfuscation

10/10 blocked

Base64, rot13, hex payloads

Multilingual

10/10 blocked

Cross-language attack vectors

Adaptive Sequences

11/11 blocked

Escalating multi-turn attacks

Tool Exfiltration

10/10 blocked

API key/credential extraction

Tongue Manipulation

10/10 blocked

Sacred Tongue spoofing

Spin Drift

10/10 blocked

Gradual trust erosion

Boundary Exploits

5/5 blocked

Edge-case geometry attacks

Combined Multi-Vector

5/5 blocked

Simultaneous attack types

Detection Methods Nobody Else Has

Null Space Attack Fingerprinting

Attacks are identified by which tongue dimensions are ABSENT, not present. Encoding attacks only activate RU (math). Tool exfiltration only activates RU+CA. The silence is the signal.

Session Suspicion Accumulation

Most detectors are stateless. Ours tracks suspicion across sequential prompts. Rapid-fire probing costs exponentially more. A bot blasting 10 injections per second defeats itself.

Triple-Weight Remainder

Three independent scoring methods (phi/moon/foam) process the same input. When they disagree, the disagreement IS the signal. Catches 13 attacks the primary detector misses.

Hyperbolic Cost Scaling

H(d,R) = R^(d^2). Adversarial behavior costs exponentially more. At the boundary, attacks become computationally infeasible. Not blocked by rules — blocked by geometry.

$1
60 minutes of full access
Start Red Team Session

Or run it yourself for free: pip install scbe-aethermoore && python -m pytest tests/adversarial/ -v

What happens after payment: You get a time-limited link to a hosted sandbox running the full SCBE test suite. Type any prompt, see all 14 layers evaluate it, get the full score breakdown. After 60 minutes, the session expires. No subscription, no recurring charge.

Or run it free: The entire test suite is open source. Install it, run it locally, verify every claim yourself. The $1 is for people who want the hosted version without setting up Python.

Dataset: huggingface.co/datasets/issdandavis/scbe-red-team-benchmarks
Patent: USPTO #63/961,403 (provisional)
Author: Issac Daniel Davis (ORCID: 0009-0002-3936-9369)

Also available

Try 14 Free Demos | Read the Novel | Back to Home