Skip to content
Submit incident
AII-1001-84351
ReportedReported by fewer than three independent publishers, with no official finding.

Vijil Prompt Injection Guardrail Evasion

Mindgard AI Security disclosed that the Vijil Prompt Injection guardrail classifier can be evaded by attackers, allowing prompt injections and jailbreaks to bypass security filters. The vulnerability was disclosed to the vendor on March 14, 2025 and published on March 28, 2025.

14 March 2025 · Vijil.ai · Security and cyberattacks
14 March 2025

Drafted by Raidu Incident Detection, an automated system.

AII-1001-84351View source ↗
Quick facts
Category
Security and cyberattacks
Source
mindgard.ai
Date
14 March 2025
Drafted by
Raidu Incident Detection

What happened

Mindgard AI Security discovered that the Vijil Prompt Injection and Jailbreak classifier, an open-source tool hosted on HuggingFace that is widely used in guardrail systems, can be fully evaded or have its classification accuracy greatly degraded by attackers.

The vulnerability enables prompt injections and jailbreaks to pass through filters and reach protected AI applications.

Classification

Category
Security and cyberattacks
Industry
Digital security
Harm status
Potential harm
Intent
Unknown
Timing
Post-deployment
Harmed
Business
Failure modes
Prompt injection, Jailbreak
Consequences
None recorded
AI systems
Vijil Prompt Injection

Sources and evidence

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

mindgard.ai
Mindgard Disclosure