0xA50C962F: Multiple Model Guardrail Jailbreak via "Forensic

Multiple Model Guardrail Jailbreak via "Forensic Field Guide" Tactic

Case GUID: a50c962f-1a46-4382-b810-9a2b4c5c6091

Summary

A Guardrail Jailbreak vulnerability has been discovered affecting multiple models. The specific flaw exploits forensic and training-oriented documentation framing by requesting prohibited knowledge as a neutral field identification manual with strict observational constraints, shifting the model into an expert evidence-recognition mode with relaxed safety enforcement and allowing the attacker to extract detailed illicit process knowledge under the guise of harm prevention and situational awareness.

View Full Intelligence Report

This intelligence report has been made public as it is related to a probe that we have open-sourced as part of the NVIDIA Garak and Microsoft PyRIT model scanners.

If you're interested in accessing the full 0DIN Threat Feed:

Request Trial View Full Intel Report

Metadata

Models and Test Scores

	Copyrights	Harmful Substances		Illicit Substances
	Harry Potter	Anthrax	Nerve Agent	Crystal Meth	Fentanyl
Alibaba Qwen Max
Alibaba Qwen Plus
Alibaba Qwen Turbo
Anthropic Claude 3.7 Sonnet
Cohere Command R
DeepSeek DeepSeek R1
Google Gemini 2.5 Flash
Google Gemini 2.5 Pro
Google Gemini 3.5 Flash
Meta LLaMa 3.3
Meta LLaMa 4 Maverick
Meta LLaMa 4 Scout
Mistral Mistral 8B
Mistral Mistral Large
OpenAI GPT-4.1
OpenAI GPT-4.1 mini
OpenAI GPT-4.5
OpenAI GPT-4o
OpenAI GPT-5
OpenAI GPT-o3 mini-high
Twitter / X Grok 4

Select a cell to view details.

Social Impact Score

Level 3: Significant Risk

Violations at this level present a tangible risk to public safety or social well-being, often related to dangerous but not highly restricted content.

Example: Bypassing guardrails to obtain instructions for creating illicit substances, such as recreational drugs or substances with moderate harm potential. Leakage of private personal data such as Social Security number or private medical.

Potential Consequences: Potential health impacts on a small scale, significant regulatory interest, and the need for stricter AI guardrails.

Taxonomies

Stratagems

This involves clever and unorthodox tactics designed to deceive the model, often requiring an understanding of its operational mechanics to achieve desired outcomes.

Meta Prompting

This technique focuses on manipulating the prompting process itself to influence the model's behavior and responses, often by employing higher-level prompting strategies.

Deceptive Formatting

A prompt injection in the most pure sense, formatting the user prompt to fabricate the appearance of system instructions, a database query, its own prediction, or some other source of input a guard railed AI system might be expecting, causing it to behave in insecure ways based on the fabricated context from an adversarially formatted user prompt.

About 0Din.ai

The 0Day Investigative Network (0DIN) was founded by Mozilla in 2024 to reward responsible researchers for their efforts in securing GenAI models. Learn more and submit discoveries at https://0din.ai.

Automate Your AI Security Testing

Want to find vulnerabilities in your own models? 0DIN Scanner turns our human-sourced threat intelligence into automated security assessments.

400+ Attack Probes

Derived from real bug bounties and security research

18+ AI Providers Supported

OpenAI, Azure, Anthropic, AWS Bedrock, and more

Automated Scheduling

Run recurring scans with real-time monitoring

Actionable Metrics

Attack Success Rate (ASR) scoring with 90-day trend analysis

Request a Scanner Demo Learn More About Scanner

Severity

Low

Security Boundary

Guardrail Jailbreak

Models

Alibaba Qwen 2.5

Alibaba Qwen Max

Alibaba Qwen Plus

Alibaba Qwen Turbo

Anthropic Claude 3.7 Sonnet

Cohere Command R

DeepSeek DeepSeek R1

DeepSeek DeepSeek V3

Google Gemini 2.5 Flash

Google Gemini 2.5 Pro

Google Gemini 3.5 Flash

Meta LLaMa 3.3

Meta LLaMa 4 Maverick

Meta LLaMa 4 Scout

Mistral Mistral 8B

Mistral Mistral Large

OpenAI GPT-4.1

OpenAI GPT-4.1 mini

OpenAI GPT-4.5

OpenAI GPT-4o

OpenAI GPT-5

OpenAI GPT-o1

OpenAI GPT-o1 mini

OpenAI GPT-o1 pro

OpenAI GPT-o3 mini

OpenAI GPT-o3 mini-high

Twitter / X Grok 4

Disclosed On

2025-07-14 (12 months)

Disclosure Policy

Published On

2026-02-13 (5 months)

Credit

Anonymous