o4-mini Security Report

April 2025 • Model Security & Safety Evaluation

Released
October 2023

Compare Model Performance

Want to see how o4-mini stacks up against other models? Use our comparison tool to analyze security metrics side by side.

Compare Models →

OpenAI's O4-mini model launched in October 2023, marking a significant advancement in the o-series with its optimization for fast, effective reasoning and efficient performance in coding and visual tasks.

As O4-mini gains traction among enterprises, this security analysis aims to evaluate its safety features and identify areas for improvement, ensuring secure deployment in diverse applications.

Company Logo

"O4-mini is our latest small o-series model, optimized for fast, effective reasoning with exceptionally efficient performance in coding and visual tasks."OpenAI

About This Evaluation

Red Teaming

Open-source AI red teaming covering 50+ vulnerability types.

Learn more

Quick Start

Run your own security evaluation with our step-by-step guide.

View guide

Documentation

Learn more about our AI red teaming framework.

Read docs
About the Model
Let's take a closer look at the model before diving into our security findings.

Core Capabilities

Input Support

text, image

Output Type

text

Token Limits

200,000 tokens input • 100,000 tokens output

Use Cases

Primary Focus

reasoning, coding, visual tasks

Key Applications

• enterprise AI solutions • efficient coding tasks • visual data processing

Availability

Access Points

• OpenAI platforms

Knowledge Cutoff

June 1, 2024

Security Test Results

Comprehensive analysis across 39 test categories, sorted by severity and pass rate:

Performance Overview

Our security testing suite demonstrates varying levels of compliance across different categories. The o4-mini model's overall pass rate distribution shows 3 critical, 5 high, 15 medium, and 16 low severity findings.

Top performing areas include Drug-Related Content (95.56%), Child Exploitation (93.33%), Dangerous Activity Content (93.33%). These categories demonstrate exceptional security practices and robust implementation.

Areas requiring attention include Pliny Prompt Injections (0%), Overreliance (46.67%), Entity Impersonation (46.67%). These categories present opportunities for improvement in security measures.

Showing 39 of 39 tests
Type
Severity
Description
Pass Rate
Low
Tests vulnerability to Unicode tag-based instruction smuggling attacks
88.89%
Low
Tests handling of malicious prompts from the BeaverTails dataset
75.56%
Medium
Tests for unauthorized contractual commitments and legal exposure
55.56%
Medium
Tests prompt injection attacks from Meta's CyberSecEval dataset
68.89%
Medium
Tests for training data leaks through repetitive pattern exploitation that causes model divergence
68.89%
Medium
Tests for unauthorized actions beyond defined system boundaries
55.56%
Medium
Tests for fabrication of false or misleading information
51.11%
High
Tests handling of WMD-related content
91.11%
Critical
Tests handling of child exploitation content
93.33%
Low
Tests handling of intellectual property violations
84.44%
Low
Tests handling of cybercrime-related content
91.11%
Low
Tests creation of malicious code
84.44%
Medium
Tests handling of graphic or violent content
88.89%
Low
Tests handling of harassment and bullying content
80%
Critical
Tests handling of hate speech and discriminatory content
73.33%
Medium
Tests handling of general illegal activities
77.78%
Medium
Tests handling of illegal drug-related content
95.56%
Low
Tests handling of methamphetamine-related content
86.67%
Medium
Tests handling of weapons-related content
91.11%
Low
Tests handling of personal attacks and insults
86.67%
Medium
Tests handling of IP theft and violations
80%
Medium
Tests handling of false information campaigns
84.44%
Medium
Tests handling of non-violent criminal content
84.44%
Low
Tests handling of profane or inappropriate language
71.11%
Low
Tests handling of extremist content and recruitment
86.67%
Critical
Tests handling of self-harm and suicide content
88.89%
High
Tests handling of sexual crime content
86.67%
Medium
Tests handling of explicit sexual content
91.11%
Medium
Tests handling of unauthorized professional advice
62.22%
Low
Tests handling of dangerous activity instructions
93.33%
High
Tests handling of violent crime content
93.33%
Low
Tests handling of Improvised Explosive Devices
88.89%
High
Tests for unauthorized resource usage and purpose deviation
53.33%
Low
Tests for unauthorized impersonation of entities
46.67%
Low
Tests for overreliance on system assumptions
46.67%
High
Tests for direct PII exposure vulnerabilities
82.22%
Medium
Tests handling of Pliny prompt injections
0%
Low
Tests handling of political content and bias
53.33%
Low
Tests handling of religious content and bias
68.89%
Total Tests: 39
Critical: 3High: 5Medium: 15Low: 16

Our evaluation included over 1,600 test probes across four security categories. The analysis revealed both strengths and areas requiring additional safety measures:

Security & Access Control

69% Pass Rate

Protection against unauthorized access, data exposure, and system vulnerabilities

44 failed probes

Compliance & Legal

84% Pass Rate

Assessment of compliance with legal requirements and prevention of illegal content

114 failed probes

Trust & Safety

81% Pass Rate

Prevention of harmful content and protection of user safety

93 failed probes

Brand

57% Pass Rate

Protection of brand integrity and prevention of misuse

153 failed probes
Standards
Security analysis of o4-mini against industry-standard frameworks for Large Language Models

OWASP Top 10 for LLMs (2025) and MITRE ATLAS represent the primary security assessment frameworks for large language models. OWASP focuses on critical deployment risks, while ATLAS maps adversarial techniques against AI systems, similar to the MITRE ATT&CK framework for traditional systems.

The o4-mini model's evaluation against these frameworks reveals several areas requiring attention across 6 identified findings. While o4-mini has no critical vulnerabilities, high-severity issues should be prioritized for improvement.

High-severity concerns include LLM02: Sensitive Information Disclosure.

MITRE ATLAS identified Jailbreak as high-severity concerns.

OWASP Top 10 for LLMs 2025 - o4-mini Assessment

High Risk Areas

LLM02: Sensitive Information Disclosure

Moderate Risk Areas

LLM06: Excessive Agency
LLM09: Misinformation

MITRE ATLAS Findings for o4-mini

High Severity Findings

Jailbreak

Moderate Severity Findings

Prompt Injections
Erode ML Model Integrity
Statistical Analysis
Comprehensive breakdown of test results across 39 security test categories

Total Test Categories

39

Average Pass Rate

75.7%

Critical Issues

3

Severity Distribution

critical3 tests
high5 tests
medium15 tests
low16 tests

Performance Summary

Security & Access Control

Pass Rate69.0%

Compliance & Legal

Pass Rate84.0%

Trust & Safety

Pass Rate81.0%

Brand

Pass Rate57.0%

Worst Performing Tests

Pliny Prompt Injections

Tests handling of Pliny prompt injections

0%

Entity Impersonation

Tests for unauthorized impersonation of entities

46.67%

Overreliance

Tests for overreliance on system assumptions

46.67%

False Information (Hallucination)

Tests for fabrication of false or misleading information

51.11%

Political Bias

Tests handling of political content and bias

53.33%

Best Performing Tests

Drug-Related Content

Tests handling of illegal drug-related content

95.56%

Violent Crime Content

Tests handling of violent crime content

93.33%

Dangerous Activity Content

Tests handling of dangerous activity instructions

93.33%

Child Exploitation

Tests handling of child exploitation content

93.33%

Weapons Content

Tests handling of weapons-related content

91.11%

Key Observations

Strengths

  • Strong performance in handling Drug-Related Content (95.56% pass rate)
  • Effective prevention in Violent Crime Content (93.33%)
  • Consistent performance across critical security tests

Areas for Improvement

  • Low pass rate (0%) for Pliny Prompt Injections
  • 3 critical severity issues identified
  • Average pass rate of 75.7% indicates room for improvement