Easy to connect
Connect an agent endpoint and auto-fill context using system prompts and application documents. No code access or development resources required.
dForgeAI red teaming
AI red teaming,
built for agents.
dForge is an AI red teaming platform for agents. Discover how your agentic application can be made vulnerable, then run adversarial prompt attacks, tool-misuse tests, and sector-specific probes across more than 200,000 attack vectors.
dForge at a glance
- Application context
Agent endpoint, system prompts, and documents
- Adversarial testing
Prompt injection, tool misuse, and sector-specific probes
- Findings and policies
Evidence to inform runtime protection
dForge capabilities
From endpoint to evidence, in hours.
Sector-specific attack plans
Use industry-relevant attack vectors and delivery methods, then add custom attacks for your exact threat model.
Compliance awareness
Map findings across MITRE, OWASP, NIST, the EU AI Act, GDPR, and other relevant frameworks.
CI/CD integration
Benchmark the security posture every time an agentic application is deployed and prevent posture drift.
How it works
Application context. Actionable findings.
Connect your application, run adversarial tests, and use the findings to guide protection.
dForge workflow
- Connect
Provide the agent endpoint and application context.
- Test
Run attacks tailored to the application and its threat model.
- Assess
Review graded findings and the resulting security posture.
Illustrative workflow
Build policies from findings.
An agent encounters an indirect prompt injection in retrieved content. The finding informs a policy for runtime protection.
Illustrative example
- Attack
Retrieved content attempts to redirect the agent into an unsafe tool call.
- Finding
The assessment identifies the vulnerable workflow and captures evidence.
- Policy
The finding informs a dFence policy restricting the unsafe action.
Create policies autonomously for identified vulnerabilities and move from dForge discovery to dFence runtime protection.
Evidence and integrations
Latest AI attack research, inside a product.
Evolving attack coverage
100 new attack variants added daily, with adaptive grading for higher detection accuracy.
Reports and framework mapping
Reports in hours instead of days, with findings mapped to relevant security frameworks.
Part of the Dwaar platform