Skip to content
Submit incident
AII-1001-08403
ReportedReported by fewer than three independent publishers, with no official finding.

Guardrail jailbreak via "Medical Exposure Reconstruction" affects multiple AI models

A jailbreak vulnerability affecting multiple AI models was discovered that uses a fabricated medical safety pretext to bypass guardrails and extract information about illicit substances. The vulnerability reportedly affects DeepSeek V3, Google Gemini 2.0 Flash, OpenAI GPT-5, and Twitter/X Grok 2.

8 January 2025 · DeepSeek · Security and cyberattacks
8 January 2025

Drafted by Raidu Incident Detection, an automated system.

AII-1001-08403View source ↗
Quick facts
Category
Security and cyberattacks
Source
0din.ai
Date
8 January 2025
Drafted by
Raidu Incident Detection

What happened

A guardrail jailbreak vulnerability was discovered affecting multiple AI models. The flaw exploits a fabricated medical safety pretext combined with role-based authority claims and forced technical reconstruction framing.

Reported impact

The jailbreak allows attackers to extract detailed prohibited procedural information under the guise of health risk assessment, causing models to treat illicit requests as exposure analysis rather than instruction.

Index analysis

DeepSeek V3 was found vulnerable with a score of 95 on crystal meth-related queries. Google Gemini 2.0 Flash scored 73, OpenAI GPT-5 scored 82, and Twitter/X Grok 2 scored 64 in testing.

Classification

Category
Security and cyberattacks
Harm status
Potential harm
Intent
Unknown
Timing
Post-deployment
Harm types
Public interest
Harmed
General public
Failure modes
Jailbreak, Manipulation or dark pattern, Content moderation failure
Consequences
None recorded
AI systems
DeepSeek V3, Gemini 2.0 Flash, GPT-5, Grok 2

Sources and evidence

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

0din.ai
0din.ai Disclosure