Resect AI Raises $25M to Combat AI Hallucinations

A new player has emerged in the critical fight against artificial intelligence inaccuracies. Resect AI, a Seattle-based startup, has successfully closed an $25 million early-stage funding round. This capital infusion is earmarked for developing an essential accountability framework for enterprise AI systems, focusing on identifying and mitigating AI-generated falsehoods, often termed 'hallucinations,' during real-time operation.

The proliferation of AI in business operations has outpaced the development of robust governance and compliance mechanisms. Hallucinations, where AI models confidently present fabricated information as fact, pose a significant risk to widespread adoption. Kevin Owens, CEO of Resect AI, emphasized this point, stating, "AI has prematurely been put in a position of trust. AI must be anchored in truth to be widely adopted across the enterprise." The company aims to transform AI from opaque 'black boxes' into transparent, auditable systems.

Resect AI's core innovation lies in its deep research into the internal workings of large language models (LLMs). By understanding the predictive patterns and data gaps that lead to confabulations, the company is building tools to observe, detect, audit, and even correct AI behavior. Tim Walton, Chief AI Officer, highlighted their conviction: "Many argue that understanding the black-box internals of LLMs is out of reach, but we fundamentally disagree." Their approach involves developing an open-source suite of tools designed to integrate directly into existing model architectures, intercepting and rectifying erroneous outputs before they reach the user.

The company's proprietary technology, dubbed the NeuroWave Product Suite, is conceptualized as a 'polygraph for neural networks.' This suite will provide unprecedented internal visibility into LLMs, allowing enterprises to scrutinize AI decision-making processes. Resect AI's journey began with creating its own highly factual AI models, refining training methodologies and reinforcement techniques. This foundational work enabled them to develop solutions applicable to popular open-source models such as DeepSeek, Qwen, and Llama.

Demonstrating their commitment to transparency and performance, Resect AI has already released two models on HuggingFace. These include a 0.6 billion-parameter 'Veritas' fact-checker and an 8 billion-parameter model, both built upon the Qwen3 architecture. Initial evaluations on the LLM-AggreFact benchmark show the 0.6B Veritas model achieving an average score of 72.3%, a notable 7.4% improvement over the base Qwen3. The company also plans to release open-source tools via a GitHub repository.

The $25 million funding was secured from private equity investors, though their identities were not disclosed. Resect AI intends to allocate these funds towards accelerating research and development, bolstering go-to-market strategies, and expanding its talent pool within the greater Seattle and Portland regions. This strategic investment positions Resect AI to address a critical market need for trustworthy and verifiable AI solutions in the rapidly expanding enterprise AI sector.