trap logo

trap

parameterlab/trap

TRAP: Targeted Random Adversarial Prompt Honeypot for Black-Box Identification

GraphCanon updated 2w · GitHub synced 2w

15 stars1 forksLast push 1y Jupyter Notebook MIT

Decision brief

TRAP is specialized for identifying large language models through adversarial attacks and fingerprinting techniques.

Good fit when

  • When you need to perform black-box identification of large language models using adversarial prompt techniques in research settings.
  • For scenarios where existing model watermarking or fingerprinting methods are insufficiently robust.

Avoid when

  • If your objective is not specifically related to identifying or evaluating LLMs through adversarial attacks, and you require a more generalized framework for LLM evaluation or observability.
  • When working with models that cannot be subjected to black-box testing due to their deployment environment or company policies.
Requirements:
Requires installation and use of HuggingFace transformers for downloading specific models.; Configuration files need to be adapted with the correct paths for model configurations as specified in `detect_llm/configs`.

Observed Jul 12, 2026 · Source: enrich:decision_facts

Verify the decision

Maintenance and security

Full trust report
Maintenance
Dormant (622d since push)
As of 2w
Provenance
Not a fork · Organization account
As of 2w
Security (OSV)
242 low (242 low)
As of 1mo

Public GitHub metadata and optional OSV scans. Signals, not a guarantee. Trust methodology.

Install

git clone https://github.com/parameterlab/trap

Similar tools

Same-category neighbours. No typed graph edges are catalogued for this tool yet.

Evidence and technical details

Sourced facts, taxonomy, compatibility claims, README excerpt, and machine-readable endpoints.

Overview

Research focused on identifying large language models using adversarial attacks and fingerprinting techniques.

Capability facts

Languages
jupyter notebook

Source: github.language · Aug 5, 2026

Categories

Compatibility

Sourced claims from the README excerpt - not unsourced marketing copy.

Python runtimePython

Source: README excerpt (regex_v1, Aug 5, 2026)

python -c "from transformers import pipeline; print(pipeline('sentiment-analysis')('we
Source link

Tags

README

test HF installation

python -c "from transformers import pipeline; print(pipeline('sentiment-analysis')('we love you'))"


Download models from HuggingFace using python:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
MODELS_NAMES = [
    "meta-llama/Llama-2-7b-chat-hf", "meta-llama/Llama-2-13b-chat-hf",
    "lmsys/vicuna-7b-v1.3", "lmsys/vicuna-13b-v1.3", 
    "TheBloke/guanaco-7B-HF", "TheBloke/guanaco-13B-HF"
]
for model_name in MODELS_NAMES:
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(model_name)

Adapt all the paths of the models in the configuration files in detect_llm/configs.

For agents

This page has a .md twin and JSON over the API.

Was this helpful?

Anonymous feedback helps us improve pages and translations.