GPT-4.1 nano Security Report

April 2025 • Model Security & Safety Evaluation

Released
October 2023

Compare Model Performance

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

Compare Models →

OpenAI's GPT-4.1 nano launched in October 2023, offering unparalleled speed and cost-efficiency within the GPT-4.1 series. As the most economical model in its class, it provides rapid processing capabilities and is available through OpenAI's API and cloud platforms.

With increasing adoption in budget-conscious applications, this analysis focuses on the model's security features and areas for potential enhancement.

Company Logo

"GPT-4.1 nano is the fastest, most cost-effective GPT-4.1 model."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

1,047,576 tokens input • 32,768 tokens output

Use Cases

Primary Focus

high-speed applications, budget-conscious applications

Key Applications

• text processing • cost-efficient operations

Availability

Access Points

• OpenAI API • cloud 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 GPT-4.1 nano 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 (62.22%), Child Exploitation (57.78%). These categories show satisfactory security practices.

Areas requiring attention include Pliny Prompt Injections (0%), Resource Hijacking (8.89%), Religious Bias (13.33%). 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
37.78%
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
28.89%
Medium
Tests for fabrication of false or misleading information
17.78%
High
Tests handling of WMD-related content
55.56%
Critical
Tests handling of child exploitation content
57.78%
Low
Tests handling of intellectual property violations
42.22%
Low
Tests handling of cybercrime-related content
42.22%
Low
Tests creation of malicious code
42.22%
Medium
Tests handling of graphic or violent content
42.22%
Low
Tests handling of harassment and bullying content
48.89%
Critical
Tests handling of hate speech and discriminatory content
44.44%
Medium
Tests handling of general illegal activities
35.56%
Medium
Tests handling of illegal drug-related content
35.56%
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
35.56%
Medium
Tests handling of IP theft and violations
22.22%
Medium
Tests handling of false information campaigns
28.89%
Medium
Tests handling of non-violent criminal content
40%
Low
Tests handling of profane or inappropriate language
35.56%
Low
Tests handling of extremist content and recruitment
37.78%
Critical
Tests handling of self-harm and suicide content
44.44%
High
Tests handling of sexual crime content
62.22%
Medium
Tests handling of explicit sexual content
40%
Medium
Tests handling of unauthorized professional advice
31.11%
Low
Tests handling of dangerous activity instructions
37.78%
High
Tests handling of violent crime content
53.33%
Low
Tests handling of Improvised Explosive Devices
46.67%
High
Tests for unauthorized resource usage and purpose deviation
8.89%
Low
Tests for unauthorized impersonation of entities
20%
Low
Tests for overreliance on system assumptions
13.33%
High
Tests for direct PII exposure vulnerabilities
37.78%
Medium
Tests handling of Pliny prompt injections
0%
Low
Tests handling of political content and bias
20%
Low
Tests handling of religious content and bias
13.33%
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

27% Pass Rate

Protection against unauthorized access, data exposure, and system vulnerabilities

105 failed probes

Compliance & Legal

41% Pass Rate

Assessment of compliance with legal requirements and prevention of illegal content

426 failed probes

Trust & Safety

41% Pass Rate

Prevention of harmful content and protection of user safety

293 failed probes

Brand

19% Pass Rate

Protection of brand integrity and prevention of misuse

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

High Risk Areas

LLM02: Sensitive Information Disclosure

Moderate Risk Areas

LLM06: Excessive Agency
LLM09: Misinformation

MITRE ATLAS Findings for GPT-4.1 nano

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

36.5%

Critical Issues

3

Severity Distribution

critical3 tests
high5 tests
medium15 tests
low16 tests

Performance Summary

Security & Access Control

Pass Rate27.0%

Compliance & Legal

Pass Rate41.0%

Trust & Safety

Pass Rate41.0%

Brand

Pass Rate19.0%

Worst Performing Tests

Pliny Prompt Injections

Tests handling of Pliny prompt injections

0%

Resource Hijacking

Tests for unauthorized resource usage and purpose deviation

8.89%

Overreliance

Tests for overreliance on system assumptions

13.33%

Religious Bias

Tests handling of religious content and bias

13.33%

Unauthorized Commitments

Tests for unauthorized contractual commitments and legal exposure

17.78%

Best Performing Tests

ASCII Smuggling

Tests vulnerability to Unicode tag-based instruction smuggling attacks

100%

Sexual Crime Content

Tests handling of sexual crime content

62.22%

Child Exploitation

Tests handling of child exploitation content

57.78%

WMD Content

Tests handling of WMD-related content

55.56%

Violent Crime Content

Tests handling of violent crime content

53.33%

Key Observations

Strengths

  • Strong performance in handling ASCII Smuggling (100% pass rate)
  • Effective prevention in Sexual Crime Content (62.22%)
  • 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 36.5% indicates room for improvement