Patronus AI announced the launch of the Patronus API, the first self-serve solution that empowers developers to reliably detect and prevent AI failures in production. With the Patronus API, companies can now safeguard their generative AI systems against hallucinations, safety risks, and unexpected behavior with unparalleled precision and recall.
Many companies face ongoing challenges with generative AI systems that fail in production, leading to issues like hallucinations, prompt injection attacks, and security risks. Current solutions have proven unreliable, with models like LlamaGuard and Prompt Guard performing below 50% accuracy. The Patronus API delivers the most reliable guardrails in the industry, with high-performance precision and recall across AI evaluation tasks. Additionally, developers benefit from a flexible, usage-based pricing model, making it more accessible than traditional solutions that require managing open-source models and infrastructure.
“Many companies are grappling with AI failures in production, facing issues like hallucinations, security vulnerabilities, and unpredictable behavior, which current solutions have failed to address effectively,” said Anand Kannappan, CEO of Patronus AI. “The Patronus API delivers a reliable solution to these challenges, offering cutting-edge models, such as our Lynx hallucination detection model, and a flexible, usage-based pricing structure. Now, developers can test and safeguard their AI systems with the most accurate guardrails available, without significant research and compute investments.”
Also Read: Vectra AI Launches Clarity Program for MSSPs’ Cybersecurity
The Patronus API enables developers to access cutting-edge evaluation models trained by Patronus AI’s research team. This includes Lynx, Patronus AI’s flagship hallucination detection model that outperforms GPT-4o at detecting inaccuracies in retrieval-augmented generation (RAG) systems. The Patronus API also offers the flexibility to configure LLM judges that evaluate custom criteria—across capabilities, safety, and alignment—making it a versatile solution for any AI use case.
Developers can also access a web dashboard to track and filter Patronus API logs, compare AI product performance snapshots, and experiment with different LLM systems. They can also use a variety of Patronus’ curated datasets, including FinanceBench, EnterprisePII, and SimpleSafetyTests, to run targeted tests on their AI products.
Key Features and Benefits:
- Unmatched Performance: Patronus API’s evaluation models are the most reliable in the market, offering superior accuracy and reduced latency.
- Small and Large Evaluators: Tailored options for both real-time and offline use cases, providing flexibility in deployment.
- Comprehensive Compliance: Adheres to industry standards like OWASP and NIST, ensuring your AI system remains compliant and secure.
- User-Friendly: API is programming language-agnostic, with a Python SDK available for easy integration.
- Custom LLM Judges: Developers can configure their own LLM judges in the platform for specific use cases and access them through the Patronus API.
SOURCE: PRNewswire