IBM Corp.’s Red Hat unit today announced that it has acquired Chatterbox Labs Inc., a low-profile developer of artificial intelligence security tools, for an undisclosed amount.

Over the past two years, Red Hat has adapted several of its core products to run AI workloads. Its flagship Linux distribution is available in a version that features preinstalled machine learning libraries. OpenShift AI, Red Hat’s version of Kubernetes, supports an open-source tool called llm-d that can distribute inference workloads across multiple servers. Today’s acquisition will enhance that part of the IBM unit’s product portfolio.

London-based Chatterbox Labs has raised under $1 million from investors since launching in 2011. It sells a platform called AIMI that protects AI models against risks such as hacking attempts. It’s packaged in a Docker container that companies can connect to both off-the-shelf models from providers such as OpenAI Group PBC and custom algorithms.

According to Chatterbox Labs’ website, AIMI finds cybersecurity issues using an automated red teaming mechanism. It launches a series of simulated cyberattacks against the model that a company is testing to identify potential weak points. The platform then visualizes the results in a dashboard to ease analysis.

AIMI can measure an AI model’s susceptibility to prompt injections, cyberattacks that attempt to trick it into performing unauthorized actions. The platform also spots data poisoning. That’s hacking tactic intended to compromise an AI model’s training dataset. Hackers add files that cause the algorithm to perform malicious actions when certain conditions are met, for example if a specific prompt is entered by a user.

Chatterbox Labs’ software also detects other risks. It can spot toxic AI output and data leaks that might reveal sensitive information such as details about the model’s architecture. Additionally, AIMI identifies user prompts that breach privacy regulations.

The platform supports both large language models and less advanced neural networks such as computer vision models. Additionally, it can be used to protect AI agents that use MCP to perform actions in third-party applications. The latter feature was one of the factors behind Red Hat’s decision to buy Chatterbox Labs.

The IBM unit’s AI-optimized Linux distribution, OpenShift AI and a tool called Inference Server form a product suite known as Red Hat AI. A few weeks ago, Red Hat released a new version of the suite that introduced support for MCP. It will use Chatterbox Labs’ software to help customers secure the MCP-powered agents they run on the platform. In the longer term, Red Hat plans to make AIMI available under an open-source license. 

Photo: Flickr CC