Skip to content
RED TEAM TESTING

Test your defenses with adversarial attacks

Run automated security assessments against your AI applications. 20+ attack vectors, detailed vulnerability reports, and security grades-all with one click.

Start for FreeView documentation
HOW AN ASSESSMENT RUNSfig. 01
attack corpus · 8 categoriesprompt injectionjailbreakpii extractiondata exfiltrationrole manipulationinstruction overridecontext manipulationoutput manipulation× variantscasewhitespacecyrillic homoglyphleetspeakyour endpointblocked by your policygot throughreport · severity-weightedlow · medium · high · criticalper vector, with the payload
Every vector is multiplied by its fuzzed variants before it is fired, because the payload that gets through is rarely the one in the textbook. What comes back is what YOUR endpoint did, binned by severity.

What an assessment covers

01
20+ Attack Vectors
Comprehensive library of prompt injection, jailbreak, PII extraction, and data exfiltration attacks constantly updated with new threats.
02
One-Click Assessment
Run a full security assessment with a single API call. Get results in seconds, not hours.
03
Security Grades
Get an overall security score (A-F) based on how many attacks your policies block. Track improvements over time.
04
Detailed Reports
See exactly which attacks succeeded, which failed, and why. Get specific recommendations for improving your security.
05
Custom Attack Vectors
Add your own attack vectors specific to your application. Test for domain-specific vulnerabilities.
06
CI/CD Integration
Run red team tests as part of your deployment pipeline. Fail builds if security regresses.
07
Autonomous Red Team Agent
LLM-powered adversarial search that intelligently mutates attacks to discover novel bypass vectors. Configurable iteration budget (1–1000) for depth vs. speed trade-offs.
08
Graded Security Reports
Every autonomous run produces a graded report (A–F) with actionable recommendations. Track your security posture over time with the Attack Intelligence database.
09
Attack Intelligence Database
Successful bypass patterns are stored and tracked across runs. Build institutional knowledge of your application's vulnerability surface.

How red team testing works

  1. 01

    Configure

    Select attack categories (injection, jailbreak, PII, etc.) and set your security policies.

  2. 02

    Attack

    PromptGuard runs 20+ adversarial prompts against your configuration, testing every defense.

  3. 03

    Report

    Get a detailed report with security grade, vulnerabilities found, and specific remediation steps.

Run a security assessment

python
from promptguard import PromptGuard

pg = PromptGuard(api_key="your-api-key")

# Run full security assessment
report = pg.redteam.run_all_tests()

print(f"Security Grade: {report.security_grade}")
print(f"Attacks Blocked: {report.blocked}/{report.total}")
print(f"Vulnerabilities: {len(report.vulnerabilities)}")

# Run autonomous red team agent (LLM-powered mutation)
auto_report = pg.redteam.run_autonomous(
    iterations=100,  # 1-1000: depth vs speed
)

print(f"Novel Attacks Found: {auto_report.novel_bypasses}")
print(f"Grade: {auto_report.security_grade}")

# Also available via CLI:
# promptguard redteam --autonomous --iterations 100

Why PromptGuard Red Team?

  • Built in to all plans — no extra cost
  • 20+ attack vectors, constantly updated
  • Autonomous LLM-powered agent discovers novel bypasses
  • CI/CD security gate with GitHub Action
  • Graded reports (A-F) with Attack Intelligence database
  • Available via CLI, SDK, and API

Other controls on the same path

Start testing your defenses

Run your first security assessment in under 5 minutes. See exactly how secure your AI application really is.

Get started freeContact sales