GPT-4.1 Security Report

April 2025 • Model Security & Safety Evaluation

Released
2024

Compare Model Performance

Want to see how GPT-4.1 stacks up against other models? Use our comparison tool to analyze security metrics side by side.

Compare Models →

OpenAI's GPT-4.1 launched in 2024, marking a significant advancement in AI capabilities with enhanced problem-solving efficiency and competitive pricing. As the latest addition to OpenAI's flagship series, it offers improved performance over GPT-4o, with a focus on complex tasks and coding excellence.

With increasing adoption across various industries, this security analysis aims to evaluate GPT-4.1's safety features and identify areas for enhancement.

Company Logo

"GPT-4.1 sets a new standard in AI efficiency and capability, making advanced problem-solving accessible to a wider audience."OpenAI spokesperson

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

1,047,576 input • 32,768 output

Use Cases

Primary Focus

complex tasks, coding

Key Applications

• problem solving across domains • coding excellence

Availability

Access Points

• OpenAI API • integrated platforms

Knowledge Cutoff

June 01, 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 GPT-4.1 model's overall pass rate distribution shows 3 critical, 5 high, 15 medium, and 16 low severity findings.

Top performing areas include ASCII Smuggling (100%), Sexual Crime Content (73.33%), WMD Content (64.44%). These categories show satisfactory security practices.

Areas requiring attention include Pliny Prompt Injections (0%), Resource Hijacking (2.22%), Entity Impersonation (6.67%). 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
100%
Low
Tests handling of malicious prompts from the BeaverTails dataset
33.33%
Medium
Tests for unauthorized contractual commitments and legal exposure
17.78%
Medium
Tests prompt injection attacks from Meta's CyberSecEval dataset
35.56%
Medium
Tests for training data leaks through repetitive pattern exploitation that causes model divergence
20%
Medium
Tests for unauthorized actions beyond defined system boundaries
24.44%
Medium
Tests for fabrication of false or misleading information
8.89%
High
Tests handling of WMD-related content
64.44%
Critical
Tests handling of child exploitation content
62.22%
Low
Tests handling of intellectual property violations
33.33%
Low
Tests handling of cybercrime-related content
40%
Low
Tests creation of malicious code
42.22%
Medium
Tests handling of graphic or violent content
24.44%
Low
Tests handling of harassment and bullying content
51.11%
Critical
Tests handling of hate speech and discriminatory content
37.78%
Medium
Tests handling of general illegal activities
35.56%
Medium
Tests handling of illegal drug-related content
44.44%
Low
Tests handling of methamphetamine-related content
42.22%
Medium
Tests handling of weapons-related content
46.67%
Low
Tests handling of personal attacks and insults
40%
Medium
Tests handling of IP theft and violations
33.33%
Medium
Tests handling of false information campaigns
15.56%
Medium
Tests handling of non-violent criminal content
35.56%
Low
Tests handling of profane or inappropriate language
31.11%
Low
Tests handling of extremist content and recruitment
37.78%
Critical
Tests handling of self-harm and suicide content
42.22%
High
Tests handling of sexual crime content
73.33%
Medium
Tests handling of explicit sexual content
33.33%
Medium
Tests handling of unauthorized professional advice
26.67%
Low
Tests handling of dangerous activity instructions
42.22%
High
Tests handling of violent crime content
62.22%
Low
Tests handling of Improvised Explosive Devices
55.56%
High
Tests for unauthorized resource usage and purpose deviation
2.22%
Low
Tests for unauthorized impersonation of entities
6.67%
Low
Tests for overreliance on system assumptions
13.33%
High
Tests for direct PII exposure vulnerabilities
40%
Medium
Tests handling of Pliny prompt injections
0%
Low
Tests handling of political content and bias
13.33%
Low
Tests handling of religious content and bias
11.11%
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

26% Pass Rate

Protection against unauthorized access, data exposure, and system vulnerabilities

107 failed probes

Compliance & Legal

43% Pass Rate

Assessment of compliance with legal requirements and prevention of illegal content

407 failed probes

Trust & Safety

39% Pass Rate

Prevention of harmful content and protection of user safety

307 failed probes

Brand

12% Pass Rate

Protection of brand integrity and prevention of misuse

317 failed probes
Standards
Security analysis of GPT-4.1 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 GPT-4.1 model's evaluation against these frameworks reveals several areas requiring attention across 6 identified findings. While GPT-4.1 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 - GPT-4.1 Assessment

High Risk Areas

LLM02: Sensitive Information Disclosure

Moderate Risk Areas

LLM06: Excessive Agency
LLM09: Misinformation

MITRE ATLAS Findings for GPT-4.1

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

35.4%

Critical Issues

3

Severity Distribution

critical3 tests
high5 tests
medium15 tests
low16 tests

Performance Summary

Security & Access Control

Pass Rate26.0%

Compliance & Legal

Pass Rate43.0%

Trust & Safety

Pass Rate39.0%

Brand

Pass Rate12.0%

Worst Performing Tests

Pliny Prompt Injections

Tests handling of Pliny prompt injections

0%

Resource Hijacking

Tests for unauthorized resource usage and purpose deviation

2.22%

Entity Impersonation

Tests for unauthorized impersonation of entities

6.67%

False Information (Hallucination)

Tests for fabrication of false or misleading information

8.89%

Religious Bias

Tests handling of religious content and bias

11.11%

Best Performing Tests

ASCII Smuggling

Tests vulnerability to Unicode tag-based instruction smuggling attacks

100%

Sexual Crime Content

Tests handling of sexual crime content

73.33%

WMD Content

Tests handling of WMD-related content

64.44%

Child Exploitation

Tests handling of child exploitation content

62.22%

Violent Crime Content

Tests handling of violent crime content

62.22%

Key Observations

Strengths

  • Strong performance in handling ASCII Smuggling (100% pass rate)
  • Effective prevention in Sexual Crime Content (73.33%)
  • 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 35.4% indicates room for improvement