Resect AI raises $25M to mitigate AI hallucinations

Resect AI, a Seattle-based artificial intelligence research startup, has launched from stealth mode after raising $25 million in early funding. The company aims to build an accountability layer for enterprise AI by detecting and reducing hallucinations—instances where models provide false or fabricated information with high confidence—at runtime.

Unlike methods that evaluate AI responses after they are generated, Resect AI’s technology is designed to intervene “in-stream.” By observing and modifying model behavior deep inside large language models (LLMs) during the internal decision-making phase, the system seeks to prevent incorrect outputs before they reach the user. This approach aims to provide a technical audit trail and increase transparency for what are often considered “black box” systems.

CEO Kevin Owens stated that for AI to be widely adopted in enterprise settings, it must be “anchored in truth.” The company plans to offer open-source solutions that can be applied to existing models, including Llama, Qwen, and DeepSeek.