Last Updated on July 25, 2024 by
Google working with Vijil to deliver trustworthy AI agents to enterprises
MENLO PARK, Calif., July 24, 2024 /PRNewswire/ — Vijil emerged from stealth today with $6 million in funding from Mayfield’s AIStart seed fund and Gradient Ventures, Google’s AI-focused seed fund. Along with funding, Vijil announced cloud services for building generative AI agents with reliability, security, and safety by design. Using Vijil cloud services to evaluate and defend trustworthy agents, AI engineers at enterprises can shorten time-to-value while maintaining governance, risk management, and compliance at all times.
Today, many organizations are drawn to the transformative potential of generative AI but are held back from deploying prototypes into production because the technology poses unknown risks to reputation and revenue. In some well-known cases, AI agents have recommended a competitor’s product, confabulated airline ticket refund policies, and concocted legal cases. Enterprises today cannot trust AI chatbots, virtual assistants, and co-pilots for business-critical use because the generative AI models inside are inherently unreliable under atypical conditions, vulnerable to attack, and prone to risk. In noisy or hostile environments, generative AI models can make egregious errors, assert falsehoods, divulge confidential or personal information, produce toxic responses, generate malware, or drive biased, unfair, unethical, or dangerous actions.
Enterprises cannot build trustworthy agents effectively because they cannot measure trust efficiently. To measure trust, AI teams either rely on external red-team consultants, resort to benchmarks, or surrender to “vibe checks”. But AI red-team consultants cannot scale to pervasive use of generative AI agents. And benchmarks are broken in several ways. Most academic benchmarks are irrelevant to enterprise use cases. Publicly available benchmark data is pulled into the training data contaminating test results. Open-source benchmarking tools are neither fast nor free – they take many days and thousands of dollars to run one benchmark. Commercial evaluation services test for task performance but fail to test reliability, security, and safety. The threats are unrelenting – AI developers must scour the media continuously for “jailbreaks” and malicious prompts. All this takes time and effort that distracts from core development. As a result, AI teams are delayed or altogether blocked from deploying AI agents that they can trust in production.
Vijil helps enterprises build and operate AI agents that people can trust. Vijil measures the trustworthiness of an AI agent by automatically evaluating its behavior with tests that are tailored to its business context. Using only a few samples of usage data from each customer, Vijil synthesizes a comprehensive test suite to measure the performance, reliability, privacy, security, and safety of the AI agent in that context. To mitigate the risks detected by the comprehensive evaluation, Vijil uses the defense-in-depth strategy to provide several layers of safeguards. The first of these is a perimeter defense mechanism that detects malicious prompts and unsafe responses with high accuracy and high speed, adaptively learning from its usage to continuously improve the AI system’s compliance with organizational policy.
“Google is pleased to collaborate with Vijil on tools to help enterprises customize Gemma and other open models for trust as well as task,” said Manvinder Singh, Director of Product Management, Google. “By adapting the Google Responsible Generative AI Toolkit to the needs of enterprises in various industries, Vijil provides critical capabilities for AI developers to preserve the privacy, security, and safety of custom models downstream with the same rigor that went into their original release.”
Using Vijil, enterprises can audit and improve trust in various generative AI systems including open large language models, closed AI APIs, Retrieval Augmented Generative (RAG) applications, and AI agents. The Vijil Evaluate cloud service provides full, fast, and easy evaluations of AI systems, offering a scalable API to execute over 1.5 million tests up to 100X faster than the next best alternative. The Vijil Dome product offers adaptive defense in real-time with the same comprehensive coverage of errors and attacks, while learning from its usage to continuously improve the accuracy and speed of detection.
“The promise of generative AI comes with the responsibility to use its potential wisely,” said Darian Shirazi, General Partner, Gradient Ventures. “Gradient is pleased to invest in Vijil to help accelerate the enterprise adoption of responsible AI with reliability, security, and safety by design.”
“Mayfield has a long history of championing entrepreneurs on their journeys from inception to iconic,” said Vijay Reddy, Mayfield Partner for AIStart Fund. “In Vijil, we saw an accomplished founding team with over a decade of experience building large-scale AI systems, a customer-obsessed approach to developing technology, and the conviction to pursue a compelling opportunity with tenacity. We envision a world where cognitive assistants work alongside humans to automate mundane tasks and complex assignments. Vijil is positioned to deliver the layer of trust in every platform built for AI agents.”
“We cannot trust autonomous agents today, no matter how intelligent they may seem, the way we trust the people we employ,” said Vin Sharma, co-founder and CEO of Vijil. “As humans, we have had 4 million years of genetic evolution and 400 thousand years of cultural evolution to understand interpersonal trust. And we have metrics and mechanisms to measure and maintain that trust. But agents must earn our trust starting with a deficit. We started Vijil to accelerate the evolution of trustworthy agents by building a trusted operating system for their development and deployment. Vijil helps you build and operate agents that you can trust based on open, safe, and secure models.”
The Vijil cloud services are available today for private preview. Enterprises can subscribe to Vijil via Google Cloud Marketplace and deploy Vijil software within their private network on Google Cloud or on-premises. Enterprises can rest assured that all application inputs, outputs, or metrics stay inside their corporate network. Vijil subscribers get continuous detection and mitigation of risks along with customized technical support. Join the waitlist at vijil.ai.
About Vijil
Founded by engineering leaders from AWS who built the LLM engines of Amazon SageMaker and Amazon Bedrock, Vijil is helping enterprises build and operate AI agents that people can trust. Backed by Mayfield AIStart and Gradient Ventures, Vijil is committed to open innovation via research collaboration with academic partners, open-source contributions to relevant projects, and alliances with industry partners who support open foundation models. Other investors and advisors include Dr. Bratin Saha, previously VP and GM for AWS AI, Joe Spisak, Director of Product for Generative AI at Meta, Dr. Leon Derczynski, Principal Research Scientist at NVIDIA and leading member of OWASP Top 10 for LLM.
SOURCE Vijil Inc
