Hirundo: Pioneering Machine Unlearning to Transform AI Reliability
In an era where artificial intelligence increasingly shapes our decisions, a startup named Hirundo is making significant strides in addressing two critical challenges: AI hallucinations and biases. This innovative company has just secured $8 million in seed funding, an investment that illustrates the urgent demand for more reliable and trustworthy AI technologies.
The Science of Forgetting: Machine Unlearning
Hirundo’s groundbreaking solution, dubbed machine unlearning, allows AI models to "forget" specific, problematic knowledge or behaviors. Unlike traditional AI techniques that merely refine or filter outputs, this method enables organizations to surgically eliminate errors and biases from already trained models. This capability means companies can address high-stakes issues without the astronomical costs and time delays associated with retraining entire models—a process that can take weeks and millions of dollars.
Picture this as AI neurosurgery: Hirundo’s technology precisely identifies and removes undesirable behaviors from an AI model’s parameters while retaining its overall efficiency—akin to performing targeted surgery rather than a full operation.
The Dangers of AI Hallucinations
AI hallucinations—instances where models produce misleading or outright false information—pose a severe threat, especially in professional environments. Research indicates that a staggering 58% to 82% of facts generated by AI for legal inquiries may be incorrect. Such inaccuracies can lead to disastrous legal repercussions and reputational harm.
Most current methods aimed at minimizing these hallucinations, like fine-tuning or implementing guardrails, often fail to tackle the root causes. In contrast, Hirundo’s approach effectively excises the harmful data that underpins these errors, providing a much-needed layer of safety for enterprises relying on AI for decision-making.
A Scalable and Flexible Solution
Hirundo’s platform is designed for versatile use across various AI applications, whether it be natural language processing, computer vision, or time-series data. It effortlessly integrates into existing systems to detect mislabeled items or data ambiguities, allowing users to backtrack and address faulty outputs— all without disrupting ongoing workflows.
For sectors handling sensitive data, like finance or healthcare, the company’s SOC-2 certified system can operate through software-as-a-service (SaaS), private clouds, or even on-premises to meet stringent compliance requirements.
Demonstrated Impact
The early performance metrics speak volumes. In tests involving popular large language models (LLMs) such as Llama and DeepSeek, Hirundo claims to have achieved:
- 55% reduction in hallucinations
- 70% decrease in bias
- 85% reduction in prompt injection attacks
These results have been validated through respected benchmarks, making Hirundo’s technology not only innovative but demonstrably effective across varied AI applications.
Founding Team’s Unique Expertise
Founded in 2023, Hirundo boasts a leadership team rich in both academic and industry experience. CEO Ben Luria, a Rhodes Scholar, co-founded notable ventures, while CTO Michael Leybovich brings awash of research credentials, alongside Chief Scientist Prof. Oded Shmueli, whose resume includes top positions in major tech firms. Their collaborative expertise equips Hirundo to tackle the complex issues currently plaguing the AI sector.
Investor Support and Future Prospects
Backed by a roster of savvy investors including Maverick Ventures Israel and SuperSeed, Hirundo stands poised to revolutionize the decision-making landscape. As noted by Yaron Carni of Maverick Ventures, the startup is addressing an urgent need for AI systems that can effectively eliminate flawed data that skews outcomes and breeds mistrust.
In a field where AI’s integration into vital infrastructure is quickly accelerating, Hirundo’s machine unlearning represents a significant leap toward safer, more reliable AI. Their technology not only addresses existing flaws but also sets a precedent for future developments tailored to trustworthiness and efficiency in AI applications.

Writes about personal finance, side hustles, gadgets, and tech innovation.
Bio: Priya specializes in making complex financial and tech topics easy to digest, with experience in fintech and consumer reviews.