{
  "id": 5592348,
  "title": "Resect launches with $25M to reduce hallucinations in AI models",
  "url": "https://urgent.news/2026/09/04/resect-launches-with-25m-to-reduce-hallucinations-in-ai-models",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-04T16:15:25.000Z",
  "source": {
    "name": "SiliconANGLE",
    "slug": "siliconangle",
    "url": "https://siliconangle.com/2026/09/04/resect-launches-with-25m-to-reduce-hallucinations-in-ai-models/"
  },
  "original_language": "en",
  "account": "Artificial intelligence startup Resect AI has raised $25 million in early funding to address the issue of hallucinations in enterprise AI models. Hallucinations occur when AI models generate false or fabricated responses, often with high confidence. Resect's core mission is to build an accountability layer that captures and reduces these hallucinations at runtime.\n\nHallucinations happen due to gaps in data or lack of real information, causing models to predict patterns based on training, not \"remembering.\" Many models are designed to appear helpful during refinement and post-training, generating wrong answers instead of admitting \"I don't know.\"\n\nResect AI aims to increase trust in AI models by anchoring them in truth. The company is developing an open-source offering for enterprise products that provides visibility into large language models, allowing for observation, detection, interpretation, audit, and modification of model behavior. Resect's open-source tools will be available on GitHub, although they are not currently populated.\n\nThe company's mission began with building AI models that produce high factuality. They developed an AI model to control data, created a training process, behavior, and reinforcement training, discovering they could be applied to existing open models such as DeepSeek, Qwen, and Llama. The focus shifted from building a better model to creating a suite of tools that work within model architecture to intercept misbehavior and redirect it before hallucinations occur.\n\nResect's NeuroWave Product Suite, a polygraph for neural networks, will form the basis of an enterprise-level audit tool. The company claims they have developed technology that observes when and how models fail and surgically fixes them. The name Resect is derived from a medical verb meaning to cut out or remove part of an organ, tissue, or bone, emphasizing their surgical approach to fixing AI models.\n\nCurrently, Resect offers two models on HuggingFace: a 0.6-billion-parameter Veritas fact checker and an 8-billion-parameter model, both based on the Qwen3 architecture and non-thinking. Performance evaluations show an improvement in the 0.6B model, scoring an average of 72.3% compared to Qwen3, an increase of 7.4%. The company plans to use the $25 million in funding for research and development, go-to-market initiatives, and hiring local talent in the Seattle and Portland regions.",
  "summary": "Seattle-based artificial intelligence research startup Resect AI announced Thursday it raised $25 million in early funding to build an accountability layer for enterprise AI by capturing and reducing hallucinations at runtime. Hallucinations, or confabulations, happen when an AI model replies with a false or fabricated response and presents it with high confidence. At their core, […] The post…",
  "key_points": [
    "Resect AI raises $25M to combat AI hallucinations",
    "Company develops open-source tools for enterprise AI models",
    "NeuroWave Product Suite aims to surgically fix model failures"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}