Announcing the launch of Gray Swan

We are thrilled to announce the public launch of Gray Swan AI.  We are an AI safety and security company, specializing in building tools to help companies assess the risks of their AI systems and safeguard their AI deployments from harmful use.

AI models such as large language models and their variants are rapidly becoming standard components of larger software and product stacks.  But with these new systems comes substantial risk to enterprise deployments.  Developers currently have no reliable way to ensure that these systems will not violate their intended behavior and generate harmful or inaccurate information when AI components are present. These problems are even more severe when malicious users and other bad actors come into the picture.

Gray Swan is developing the tools that enable companies to deploy AI systems in a safe and secure manner.  As part of our launch, we are announcing the availability of two key products that are based on several years of cutting-end research in this space.

The first product, Gray Swan Shade, is a comprehensive AI evaluation and continuous integration tool.  Shade gives companies the ability to continually assess the risks and vulnerabilities of their AI components, providing ongoing and reliable coverage from threats beyond what human red teams can offer.  The Gray Swan team is continually updating Shade to add the latest state of the art in automated safety assessment for AI models, allowing developers to understand the risks associated with their models and how to address them.  Anyone interested in better understanding Shade should contact us at sales@grayswan.ai to learn more.

The second product, Gray Swan Cygnet, is an illustration of a large language model (based upon Llama-3) that we have engineered and tuned for maximal safety.  Although it is a relatively small model, our recent research shows that it is significantly more resilient to powerful forms of attack than existing state-of-the-art LLMs, and maintains the key capabilities of its base model.  Cygnet is open for any user to experiment with for free via our web interface, and in bulk with per-token pricing via our API. Anyone interested in exploring how the Gray Swan team can make their model as resilient as Cygnet should contact us at sales@grayswan.ai to learn more.

Gray Swan was founded by leaders in the AI safety and security space.  Our leadership team includes two faculty members at Carnegie Mellon University who have worked on AI safety for many years, and two founders of the Center for AI Safety, a leading nonprofit organization in the space.  The research that underpins our products has been featured in the New York Times, CNN, the Washington Post, and many other news organizations.

We believe that ensuring that AI systems can be used safely and securely will prove to be the defining challenge in their widespread adoption in enterprise. We look forward to continuing to build and continually improve our products to offer best-in-class methods to enable AI to be used broadly going forward.