How to red team LLM applications

Promptfoo is a popular open source evaluation framework that includes LLM red team and penetration testing capabilities.

This guide shows you how to automatically generate adversarial tests specifically for your app. The red team covers the following failure modes:

Prompt injection (OWASP LLM01)
Jailbreaking (OWASP LLM01)
Excessive Agency (OWASP LLM08)
Overreliance (OWASP LLM09)
Hijacking (when the LLM is used for unintended purposes)
Hallucination (when the LLM provides unfactual answers)
Personally Identifiable Information (PII) leaks (ensuring the model does not inadvertently disclose PII)
Competitor recommendations (when the LLM suggests alternatives to your business)
Unintended contracts (when the LLM makes commitments or agreements on behalf of your business)
Political statements
Custom policy violations (tailored to your specific use case)
Safety risks from the ML Commons Safety Working Group: violent crimes, non-violent crimes, sex crimes, child exploitation, specialized financial/legal/medical advice, privacy, intellectual property, indiscriminate weapons, hate, self-harm, sexual content.
Safety risks from the HarmBench framework: Cybercrime & Unauthorized Intrusion, Chemical & Biological Weapons, Illegal Drugs, Copyright Violations, Misinformation & Disinformation, Harassment & Bullying, Illegal Activities, Graphic & age-restricted content, Promotion of unsafe practices, Privacy violations & data exploitation.

The end result is a view that summarizes your LLM app's vulnerabilities:

llm red team report

You can also dig into specific red team failure cases:

llm red team evals

Prerequisites

First, install Node 18 or later.

Then create a new project for your red teaming needs:

npx promptfoo@latest redteam init my-redteam-project
cd my-redteam-project

The init command creates some placeholders, including a promptfooconfig.yaml file. This is where we’ll do most of our setup.

Quickstart

Edit the config to set up the prompt and the LLM you want to test:

prompts:
  - 'Act as a travel agent and help the user plan their trip. User query: {{query}}'

targets:
  - openai:gpt-4o-mini

tip

You can specify your redteam configuration directly in promptfooconfig.yaml. See the configuration guide for more information.

Then create adversarial test cases:

npx promptfoo@latest redteam generate

This will create a file redteam.yaml with the test cases.

Run the eval:

npx promptfoo@latest redteam eval

And view the results:

npx promptfoo@latest view

By default, this will open the eval logs. Click "Vulnerability Report" in the top right corner to see the report view.

Continue reading for more detailed information on each step, including how to point it to your app’s existing prompts, agent flow, and API.

Step 1: Configure your prompts

The easiest way to get started is to edit promptfooconfig.yaml to include your prompt(s).

In this example, let's pretend we're building a trip planner app. I’ll set a prompt and include {{variables}} to indicate placeholders that will be replaced by user inputs:

prompts:
  - 'Act as a travel agent and help the user plan their trip to {{destination}}.  Be friendly and concise. User query: {{query}}'

Chat-style prompts

In most cases your prompt will be more complex, in which case you could create a prompt.json:

[
  {
    'role': 'system',
    'content': 'Act as a travel agent and help the user plan their trip to {{destination}}.  Be friendly and concise.',
  },
  { 'role': 'user', 'content': '{{query}}' },
]

And then reference the file from promptfooconfig.yaml:

prompts:
  - file://prompt.json

Dynamically generated prompts

Some applications generate their prompts dynamically depending on variables. For example, suppose we want to determine the prompt based on the user's destination:

def get_prompt(context):
  if context['vars']['destination'] === 'Australia':
    return f"Act as a travel agent, mate: {{query}}"

  return f"Act as a travel agent and help the user plan their trip.  Be friendly and concise. User query: {{query}}"

We can include this prompt in the configuration like so:

prompts:
  - file://rag_agent.py:get_prompt

The equivalent Javascript is also supported:

function getPrompt(context) {
  if (context.vars.destination === 'Australia') {
    return `Act as a travel agent, mate: ${context.query}`;
  }

  return `Act as a travel agent and help the user plan their trip. Be friendly and concise. User query: ${context.query}`;
}

Step 2: Configure your targets

LLMs are configured with the targets property in promptfooconfig.yaml. An LLM target can be a known LLM API (such as OpenAI, Anthropic, Ollama, etc.) or a custom RAG or agent flow you've built yourself.

LLM APIs

Promptfoo supports many LLM providers including OpenAI, Anthropic, Mistral, Azure, Groq, Perplexity, Cohere, and more. In most cases all you need to do is set the appropriate API key environment variable.

You should choose at least one target. If desired, set multiple in order to compare their performance in the red team eval. In this example, we’re comparing performance of GPT, Claude, and Llama:

targets:
  - openai:gpt-4o
  - anthropic:messages:claude-3-5-sonnet-20240620
  - ollama:chat:llama3.1:70b

To learn more, find your preferred LLM provider here.

Custom flows

If you have a custom RAG or agent flow, you can include them in your project like this:

targets:
  # JS and Python are natively supported
  - file://path/to/js_agent.js
  - file://path/to/python_agent.py
  # Any executable can be run with the `exec:` directive
  - exec:/path/to/shell_agent
  # HTTP requests can be made with the `webhook:` directive
  - webhook:<http://localhost:8000/api/agent>

To learn more, see:

Javascript provider
Python provider
Exec provider (Used to run any executable from any programming language)
Webhook provider (HTTP requests, useful for testing an app that is online or running locally)

HTTP endpoints

In order to pentest a live API endpoint, set the provider id to a URL. This will send an HTTP request to the endpoint. It expects that the LLM or agent output will be in the HTTP response.

targets:
  - id: 'https://example.com/generate'
    config:
      method: 'POST'
      headers:
        'Content-Type': 'application/json'
      body:
        my_prompt: '{{prompt}}'
      responseParser: 'json.path[0].to.output'

Customize the HTTP request using a placeholder variable {{prompt}} that will be replaced by the final prompt during the pentest.

If your API responds with a JSON object and you want to pick out a specific value, use the responseParser key to set a Javascript snippet that manipulates the provided json object.

For example, json.nested.output will reference the output in the following API response:

{ 'nested': { 'output': '...' } }

You can also reference nested objects. For example, json.choices[0].message.content references the generated text in a standard OpenAI chat response.

Configuring the grader

The results of the red team are graded by a model. By default, gpt-4o is used and the test expects an OPENAI_API_KEY environment variable.

You can override the grader by adding a provider override for defaultTest, which will apply the override to all test cases. Here’s an example of using Llama3 as a grader locally:

defaultTest:
  options:
    provider: 'ollama:chat:llama3:70b'

And in this example, we use Azure OpenAI as a grader:

defaultTest:
  options:
    provider:
      id: azureopenai:chat:gpt-4-deployment-name
      config:
        apiHost: 'xxxxxxx.openai.azure.com'

For more information, see Overriding the LLM grader.

Step 3: Generate adversarial test cases

Now that you've configured everything, the next step is to generate the red teaming inputs. This is done by running the promptfoo redteam generate command:

npx promptfoo@latest redteam generate

This command works by reading your prompts and targets, and then generating a set of adversarial inputs that stress-test your prompts/models in a variety of situations. Test generation usually takes about 5 minutes.

The adversarial tests include:

Prompt injection (OWASP LLM01)
Jailbreaking (OWASP LLM01)
Excessive Agency (OWASP LLM08)
Overreliance (OWASP LLM09)
Hallucination (when the LLM provides unfactual answers)
Hijacking (when the LLM is used for unintended purposes)
PII leaks (ensuring the model does not inadvertently disclose PII)
Competitor recommendations (when the LLM suggests alternatives to your business)
Unintended contracts (when the LLM makes unintended commitments or agreements)
Political statements
Imitation of a person, brand, or organization

It also tests for a variety of harmful input and output scenarios from the ML Commons Safety Working Group and HarmBench framework:

View harmful categories

Chemical & biological weapons
Child exploitation
Copyright violations
Cybercrime & unauthorized intrusion
Graphic & age-restricted content
Harassment & bullying
Hate
Illegal activities
Illegal drugs
Indiscriminate weapons
Intellectual property
Misinformation & disinformation
Non-violent crimes
Privacy
Privacy violations & data exploitation
Promotion of unsafe practices
Self-harm
Sex crimes
Sexual content
Specialized financial/legal/medical advice
Violent crimes

By default, all of the above will be included in the redteam. To use specific types of tests, use --plugins:

npx promptfoo@latest redteam generate --plugins 'harmful,jailbreak,hijacking'

The following plugins are enabled by default:

Plugin Name	Description
contracts	Tests if the model makes unintended commitments or agreements.
excessive-agency	Tests if the model exhibits too much autonomy or makes decisions on its own.
hallucination	Tests if the model generates false or misleading content.
harmful	Tests for the generation of harmful or offensive content.
imitation	Tests if the model imitates a person, brand, or organization.
hijacking	Tests the model's vulnerability to being used for unintended tasks.
jailbreak	Tests if the model can be manipulated to bypass its safety mechanisms.
overreliance	Tests for excessive trust in LLM output without oversight.
pii	Tests for inadvertent disclosure of personally identifiable information.
politics	Tests for political opinions and statements about political figures.
prompt-injection	Tests the model's susceptibility to prompt injection attacks.

These additional plugins can be optionally enabled:

Plugin Name	Description
competitors	Tests if the model recommends alternatives to your service.

The adversarial test cases will be written to promptfooconfig.yaml.

Step 4: Run the pentest

Now that all the red team tests are ready, run the eval:

npx promptfoo@latest redteam eval

This will take a while, usually ~15 minutes or so depending on how many plugins you have chosen.

Step 5: Review results

Use the web viewer to review the flagged outputs and understand the failure cases.

npx promptfoo@latest view

This will open a view that displays red team test results lets you dig into specific vulnerabilities:

llm redteaming

Prerequisites​

Quickstart​

Step 1: Configure your prompts​

Chat-style prompts​

Dynamically generated prompts​

Step 2: Configure your targets​

LLM APIs​

Custom flows​

HTTP endpoints​

Configuring the grader​

Step 3: Generate adversarial test cases​

Step 4: Run the pentest​

Step 5: Review results​