Protect AI Acquires Laiyer AI to Fortify Large Language Model Security

2024-02-01

Layier AI's core product, LLM Guard, aims to provide security for large language models such as GPT-4 from OpenAI LP, to address various security threats, abuses, and attacks. LLM Guard can withstand threats such as prompt injection attacks, data poisoning, and other vulnerabilities that may occur in AI models. Prompt injection attacks involve attackers manipulating LLM through carefully crafted inputs, leading to data leakage or decision manipulation. Attacks can be direct, through LLM inputs, or indirect, through contaminated data source inputs. Due to the implicit trust in LLM outputs, they often bypass detection. LLM Guard can detect, edit, and sanitize LLM inputs and outputs. This reduces the risk of manipulating model outputs leading to personal data leakage, ensuring the integrity and security of the data processed by these models. The service is optimized for performance and integrates with existing security workflows, providing observability tools such as logging and metrics. According to Layier, LLM Guard can reduce CPU inference latency to one-third of its original value, enabling cost-effective CPU usage without compromising accuracy instead of expensive GPUs. The code is open source and has gained high adoption. Within just 30 days, Layier AI has seen over 13,000 library downloads and 2.5 million model downloads on Hugging Face. Through the acquisition, Protect AI intends to create a commercial version of LLM Guard, expanding its capabilities, functionalities, and integrations within the Protect AI platform. CEO Ian Swanson stated before the announcement, "These new features will enable our customers in the automotive, energy, manufacturing, life sciences, financial services, and government sectors to develop secure and reliable GenAI applications." Layier co-founders Neal Swaelens and Oleksandr Yaremchuk added, "By partnering with Protect AI, we are expanding the product with LLM security capabilities to provide the most comprehensive end-to-end AI security platform in the industry." Protect AI is a venture-backed startup that last raised $35 million in funding in July. The company's investors include Evolution Equity Partners LLP, Salesforce Ventures LLC, Acrew Capital LP, Boldstart Ventures LLC, Knollwood Capital LLC, and Pelion Ventures Partners LLC.