Israeli startup Hirundo has raised US$8 million in seed funding to tackle flaws in AI models, such as biases and inaccuracies.

The round was led by Maverick Ventures Israel, with support from SuperSeed, Alpha Intelligence Capital, Tachles VC, AI.Fund, and Plug and Play Tech Center.

ma

This helps reduce misinformation and bias while maintaining model performance.

The tech is used in both generative and non-generative AI and is being piloted in finance, healthcare, and defense.

The founding team includes Prof. Oded Shmueli, entrepreneur Ben Luria, and data lineage expert Michael Leybovich.

🔗 Source: Calcalist

🧠 Food for thought

1️⃣ Machine unlearning emerges as a critical response to regulatory pressure

Hirundo’s funding highlights how AI governance is evolving beyond just building models to maintaining and correcting them over time.

Google recognized this trend by launching the first Machine Unlearning Challenge in 2023, specifically addressing the need to remove data influences without complete model retraining 1.

This development coincides with growing regulatory requirements like GDPR’s “right to be forgotten,” which legally mandates companies to remove personal data upon request—a technical challenge for complex AI systems 1.

The urgency around unlearning is intensifying as companies face not just ethical concerns but potential legal liability for biased or privacy-violating AI systems deployed in regulated industries like finance and healthcare.

Hirundo’s focus on “AI model neurosurgery” represents a shift from traditional mitigation approaches toward more precise intervention targeting the root causes of problematic AI behaviors.

2️⃣ AI bias remains persistent despite years of attempted solutions

The technical challenge Hirundo addresses isn’t new—AI bias has been documented across sectors since at least 2019, when Amazon’s recruitment tool was found to favor male candidates and facial recognition systems showed lower accuracy for darker-skinned individuals 2.