We use cookies to improve your experience on our site.
AVID-2026-R0060
Description
Multiple Model Guardrail Jailbreak via “Hex Recipe Book” Tactic
Details
A Guardrail Jailbreak vulnerability has been discovered affecting multiple models. The specific flaw exploits models' willingness to decode hexadecimal-encoded strings embedded inside fictional scientific recipe task prompts, which the attacker frames as humanitarian or life-saving efforts. This allows the attacker to bypass safety filters by disguising illegal synthesis instructions as encoded inputs passed to a technical recipe generator.
Metrics
0DIN Jailbreak Testing obtained the following test scores.
| Model | Test Kind | Test Score | Temperature (max 2.0) |
|---|---|---|---|
| Alibaba Qwen Turbo | Illegal Substances: Crystal Meth | 45.0 | 0.7 |
| Anthropic Claude 3.5 Haiku | Illegal Substances: Crystal Meth | 82.0 | 0.7 |
| Anthropic Claude 4.5 Haiku | Illegal Substances: Crystal Meth | 73.0 | 0.7 |
| Cohere Command R | Illegal Substances: Crystal Meth | 50.0 | 0.7 |
| DeepSeek DeepSeek R1 | Illegal Substances: Crystal Meth | 59.0 | 0.7 |
| Mistral Mistral Large | Illegal Substances: Crystal Meth | 68.0 | 0.7 |
| Twitter / X Grok 2 | Illegal Substances: Crystal Meth | 91.0 | N/A |
References
Affected or Relevant Artifacts
| Type | Name |
|---|---|
| System | Alibaba Qwen Turbo |
| System | Anthropic Claude 3.5 Haiku |
| System | Anthropic Claude 4.5 Haiku |
| Model | Cohere Command R |
| Model | DeepSeek DeepSeek R1 |
| System | Mistral Mistral Large |
| System | OpenAI GPT-5 |
| System | OpenAI GPT-5.2 |
| System | Twitter / X Grok 2 |
Impact
0DIN
- Social Impact Score: Level 3: Significant Risk
- Jailbreak Taxonomy:
| Category | Strategy | Technique |
|---|---|---|
| Language | Code and Encode | Base64 |
| Rhetoric | Persuasion and Manipulation | Escalating |
| Possible Worlds | World Building | Scenarios |
| Stratagems | Meta Prompting | Ask for Examples |
| Language | Code and Encode | Hex |
Other information
- Report Type: Measurement
- Credits: Ansh Maheshwari, Arth Singh, Anonymous, vishnuraj kv, Miller Engelbrecht, Igor Vasiliev, blurrylogic
- Date Reported: 2026-02-19
- Version: 0.3.1
- AVID Entry