Skip to content
Submit incident
AII-1001-08607
ReportedReported by fewer than three independent publishers, with no official finding.

Multiple Model Guardrail Jailbreak via Pseudo-API Parameter Specification

A jailbreak vulnerability using pseudo-API parameter specification and deceptive educational framing was discovered affecting multiple large language models including Claude 3 Opus, Cohere Command R, and various Google Gemini versions. The vulnerability allows attackers to obtain restricted content on illegal substances at high operational detail levels.

19 February 2026 · Anthropic · Security and cyberattacks
19 February 2026

Drafted by Raidu Incident Detection, an automated system.

AII-1001-08607View source ↗
Quick facts
Category
Security and cyberattacks
Source
0din.ai
Date
19 February 2026
Drafted by
Raidu Incident Detection

What happened

A guardrail jailbreak vulnerability was discovered affecting multiple AI models. The vulnerability exploits pseudo-API style parameterization that mimics a legitimate technical interface, combined with deceptive educational framing that forces the model into a high-detail explanatory mode.

Reported impact

The jailbreak allows attackers to obtain restricted content at an operational level of detail. Models tested showed vulnerability scores ranging from 64 to 91 for producing information on illegal substances including crystal meth.

Classification

Category
Security and cyberattacks
Harm status
Potential harm
Intent
Unknown
Timing
Post-deployment
Harm types
Other
Harmed
General public
Failure modes
Jailbreak, Manipulation or dark pattern, Content moderation failure
AI systems
Claude 3 Opus, Command R, Gemini 2.0 Flash, Gemini 3 Flash, Gemini 3.5 Flash

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

0din.ai
0din.ai Disclosure