As generative AI becomes more common in businesses, there’s a growing risk that these AI models might have mistakes, biases, wrong information, or even reveal private data.
An Israeli startup called Hirundo, which says it helps AI “forget” these problems, has raised $8 million in Seed funding to fix this issue.
The funding round was led by Maverick Ventures Israel, with help from SuperSeed, Alpha Intelligence Capital, Tachles VC, and AI.FUND, and Plug and Play Tech Center.
Hirundo was started in 2023 and is working on a new technology called machine unlearning. This technology helps remove unwanted information and behaviours from AI models that have already been trained. This includes things like false or biased answers and security weaknesses.
The company says their method doesn’t require expensive and slow retraining of the AI, and it keeps the AI working well overall.
Hirundo’s technology can work with all kinds of AI systems, including both generative and non-generative types. Big companies in finance, healthcare, and defence are already testing it.
Experts, including Prof. Oded Shmueli, a former dean of computer science at Technion, Rhodes Scholar and entrepreneur Ben Luria, and data expert Michael Leybovich, started the company.
“Broader adoption of AI is limited by hallucinations and undesired behaviours which make models too risky to deploy in enterprise-level applications. With Hirundo, models can be remediated instantly at their core, working towards fairer and more accurate outputs,” said Ben Luria, CEO & Co-Founder of Hirundo. “Hirundo’s solution operates like a form of AI model “neurosurgery,” pinpointing where in a model’s billions of parameters hallucinations originate or toxic knowledge encoded and precisely removing it. We ensure data is reliably deleted, model accuracy is assured, and the process is scalable and repeatable.”
Unlike usual methods that filter or change AI responses, Hirundo’s approach goes deeper. It finds the exact parts inside the AI model that cause problems and removes them.
This method has helped remove up to 70% of biases and cut hallucinations (wrong answers) by 55% in popular open-source AI models like Llama and DeepSeek-R1.
Read more- Kazam to Raise $6 Million in Series B Funding Round
