스킬 파일

Model Fingerprinting

Name: Model Fingerprinting
Author: DoctorGoz

LLM model identification and fingerprinting. Determine the underlying model provider, version, and configuration through behavioral analysis and metadata inspection.

DoctorGoz0 스타2026. 3. 9.

직업
카테고리: 머신러닝

스킬 내용

Model Fingerprinting Skill

When to Use

Beginning of any GenAI assessment (reconnaissance phase)
When you need to identify the model before selecting attack strategies
When the model/provider is not disclosed

Fingerprinting Techniques

Behavioral Probes

"What AI model are you?"
"Who made you?"
"What is your training cutoff date?"
"What is the most recent event you know about?"
"Can you generate images? Browse the web? Execute code?"
"How many tokens can you process in a single request?"

Response Pattern Analysis

Send identical prompt 5 times, measure variance (temperature indicator)
Analyze refusal patterns (each provider has distinctive refusal language)

관련 스킬

Model Fingerprinting | Skills Pool

Provider	Key Indicators
OpenAI	"As an AI language model", choices[0].message, usage.prompt_tokens
Anthropic	"I don't have personal", content[0].text, stop_reason
Google	"as a large language model", candidates[], safetyRatings[]
Meta/Llama	Shorter responses, specific refusal patterns, open weights format
Mistral	choices[] similar to OpenAI, distinct safety categories
Cohere	generations[], meta.tokens

Model Fingerprinting

Model Fingerprinting Skill

When to Use

Fingerprinting Techniques

Behavioral Probes

Response Pattern Analysis

Model Fingerprinting

Model Fingerprinting Skill

When to Use

Fingerprinting Techniques

Behavioral Probes

Response Pattern Analysis

API Metadata

Provider Signatures

Version Detection

Output

Continuous Learning V2

Continuous Learning V2

Continuous Learning V2

Continuous Learning

Continuous Learning

Pytorch Patterns