Recon is a scalable, automated red teaming solution designed for rigorous testing and vulnerability assessment of generative AI applications and large language models (LLMs). It helps organizations identify, analyze, and remediate security risks before deployment by leveraging a large attack library, AI-driven agents, and continuous threat research.

Key Features

Extensive Attack Library

• Contains over 450 known attacks across six major threat categories, updated weekly.
• Uses trained LLMs as detectors for accurate vulnerability identification and reduced false positives.
• Continuously updated to counter zero-day and emerging threats.

Precise, Relevant Red Team Attacks

• Tests the AI application, not just the model.
• Agent creates attacks based on business objectives, application logic, guardrails, RAG pipelines, and system prompts.

Augmented Red Teaming

• Red teamers can set attack goals in natural language; no coding required.
• Provides in-depth, conversation-level visibility for risk analysis and remediation.

Bring-Your-Own Attack Prompts

• Users can upload custom attack prompts to simulate specific threat scenarios.
• Proprietary detection algorithms tailor tests to unique environments.

Reporting and Framework Mapping

• Exports results to CSV and JSON for collaboration.
• Maps vulnerabilities to OWASP Top 10 for LLMs, DASF, MITRE ATLAS, and NIST AI-RMF for compliance.

Integration and Automation

• Integrates with platforms like Databricks Mosaic AI and Amazon Bedrock.
• Automated and continuous scanning options for ongoing security monitoring.

Benefits

Proactive Risk Identification

• Uncovers vulnerabilities before AI apps go into production.
• Enables data-informed risk decisions and model selection.

Accelerated Testing

• Systematically tests and retests AI apps in hours.
• Reduces the cost and time of iterative security testing.

Comprehensive Compliance

• Aligns with industry frameworks (OWASP, DASF, MITRE, NIST).
• Simplifies meeting enterprise security and governance standards.

Scalable Security

• Leverages a community of 17,000+ security researchers and academic research.
• Scales beyond human capabilities with AI-driven agents.