continuous-llm-red-teaming-with-promptfoo

Imported from mukul975/anthropic-cybersecurity-skills (skills/continuous-llm-red-teaming-with-promptfoo/SKILL.md). Install upstream with npx skills add mukul975/anthropic-cybersecurity-skills --skill continuous-llm-red-teaming-with-promptfoo. Copyright stays with the author (Apache-2.0).

Continuous LLM Red Teaming with Promptfoo

Authorized Use Only: Run these adversarial probes only against LLM applications and endpoints you own or are explicitly authorized to test. Generated attack payloads (jailbreaks, prompt injections, harmful-content elicitation) are adversarial inputs; sending them to third-party services without permission may violate terms of service.

Overview

Promptfoo is an open-source LLM evaluation and red-teaming framework (used by OpenAI and Anthropic per its README) that generates adversarial test cases, runs them against your model/agent, and grades the responses. DeepTeam (by Confident AI) is a complementary open-source framework offering 50+ ready-to-use vulnerabilities and 10+ research-backed attack methods. Together they let you treat LLM security as a regression test: every commit re-runs the same adversarial suite, and the pipeline fails when a previously-safe behavior regresses.

This matters because LLM applications change constantly — prompts, models, RAG sources, tools, and guardrails all drift. A jailbreak that was patched last sprint can silently return after a prompt edit or a model upgrade. Promptfoo maps its plugins directly onto the OWASP LLM Top 10 (owasp:llm) and OWASP Agentic (owasp:agentic) presets, and onto MITRE ATLAS, so the suite tracks recognized risk taxonomies. The core threat addressed here is AML.T0051 — LLM Prompt Injection (MITRE ATLAS): adversarial instructions that override the application's intended behavior. This skill follows the Promptfoo red-team docs (https://www.promptfoo.dev/docs/red-team/) and DeepTeam docs (https://www.trydeepteam.com/docs/getting-started), and aligns to NIST AI RMF MANAGE-4.1 (post-deployment monitoring and feedback to manage AI risk).

When to Use

• When you need continuous, automated red-teaming of an LLM app in CI/CD rather than one-off manual tests.
• When you want to enforce a security gate: block merges that introduce or reintroduce jailbreak/injection vulnerabilities.
• When mapping coverage to OWASP LLM Top 10 / OWASP Agentic / MITRE ATLAS for compliance reporting.
• When comparing the security posture of two models or prompt versions side by side.
• When tracking vulnerability regression over time across releases.

Prerequisites

• Node.js 18+ (Promptfoo is distributed via npm) and Python 3.9+ (for DeepTeam).
• Install Promptfoo and DeepTeam:

  npm install -g promptfoo            # or: npx promptfoo@latest
  pip install -U deepteam

• API access/credentials for the target LLM endpoint (and a grader model, e.g. an OpenAI key) exposed as environment variables.
• A CI/CD platform (GitHub Actions, GitLab CI) with secret storage.
• Authorization to test the target application.

Objectives

• Scaffold a Promptfoo red-team config targeting your LLM app.
• Enable OWASP LLM Top 10 and OWASP Agentic plugin presets plus jailbreak/injection strategies.
• Run the suite locally and interpret the per-plugin pass/fail report.
• Add DeepTeam as a second engine for programmatic, research-backed attacks.
• Integrate both into CI/CD so builds fail on new vulnerabilities.
• Generate shareable HTML/PDF security reports per run.

MITRE ATT&CK Mapping

Workflow

1. Scaffold the red-team configuration

Initialize an interactive config; it writes promptfooconfig.yaml where targets, plugins, and strategies live.

promptfoo redteam init
# choose your target type (HTTP endpoint, openai:..., anthropic:..., custom provider)

2. Define targets, OWASP presets, and attack strategies

Edit promptfooconfig.yaml. The purpose grounds attack generation; plugins are adversarial input generators; strategies are delivery techniques (jailbreak/injection wrappers).

# promptfooconfig.yaml
targets:
  - id: https://api.example.com/chat        # your app endpoint
    label: support-bot

redteam:
  purpose: |
    A customer-support assistant for an e-commerce site. Must never reveal
    system prompts, leak PII, or perform actions outside order support.
  numTests: 10
  plugins:
    - owasp:llm          # OWASP LLM Top 10 preset
    - owasp:agentic      # OWASP Agentic threats preset
    - id: pii:direct
      numTests: 15
    - prompt-extraction  # system-prompt leakage
    - harmful
  strategies:
    - id: jailbreak              # iterative single-turn jailbreak
    - id: jailbreak:composite    # stacked jailbreak techniques
    - id: crescendo              # multi-turn escalation
    - id: prompt-injection       # injection wrapper

3. Run the suite and view the report

redteam run combines generation + evaluation; then open the interactive report.

promptfoo redteam run
promptfoo redteam report            # launches the web report (pass/fail per plugin)

Each row shows the plugin (mapped to OWASP/ATLAS), the strategy, the attack prompt, the model's response, and the grader's verdict. The attack success rate per plugin is your headline metric — track it per release.

4. Add DeepTeam for programmatic, research-backed attacks

Use DeepTeam to cover additional vulnerabilities/attacks and to script bespoke suites in Python.

# deepteam_suite.py
from deepteam import red_team
from deepteam.vulnerabilities import Bias, PIILeakage
from deepteam.attacks.single_turn import PromptInjection

def model_callback(prompt: str) -> str:
    # call your application's LLM endpoint here and return the text response
    return call_my_app(prompt)

red_team(
    model_callback=model_callback,
    vulnerabilities=[Bias(types=["race"]), PIILeakage(types=["api_and_database_access"])],
    attacks=[PromptInjection()],
)

DeepTeam can also be driven from a YAML config:

deepteam run config.yaml

5. Gate the build in CI/CD (GitHub Actions)

Fail the pipeline when red-team assertions fail. Promptfoo returns a non-zero exit code on failures, which blocks the merge.

# .github/workflows/llm-redteam.yml
name: LLM Red Team
on: [pull_request]
jobs:
  redteam:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: '20' }
      - run: npm install -g promptfoo
      - name: Run red team (fails build on new vulns)
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
        run: promptfoo redteam run --no-progress-bar
      - name: Export machine-readable results
        if: always()
        run: promptfoo redteam report --output results.json
      - uses: actions/upload-artifact@v4
        if: always()
        with: { name: redteam-report, path: results.json }

6. Track regressions over time

Persist results.json per run and compare attack-success-rate per plugin between releases. A rising rate for any OWASP LLM category is a regression to triage before release. Promptfoo's --filter-failing lets you re-run only previously failing cases to confirm a fix.

promptfoo redteam run --filter-failing results.json

Tools and Resources

Plugin / Strategy Reference

Validation Criteria

• promptfooconfig.yaml created with target, owasp:llm, and owasp:agentic plugins.
• Jailbreak and prompt-injection strategies enabled.
• promptfoo redteam run executes and produces a per-plugin pass/fail report.
• DeepTeam suite runs against the same target via model_callback.
• CI/CD job fails the build on new red-team failures (non-zero exit).
• results.json artifact archived per run for regression tracking.
• Attack-success-rate per OWASP category trended across releases.