Inherent AI raises $50M to ask science better questions

Inherent, a London-based AI lab founded by former DeepMind and Microsoft researchers, emerged from stealth on Friday with a $50 million seed round. Index Ventures and Radical Ventures co-led the raise, with Nvidia's venture arm NVentures joining alongside Ex/Ante, Metaplanet, Macroscopic Ventures, and Mythos Ventures. The Next Web reports it ranks among Europe's largest AI stealth-to-launch rounds of 2026.

The company's core argument is that most artificial intelligence is built to answer questions, not generate them. Its platform, called Faraday, pairs human scientists with self-improving AI agents designed to identify which scientific problems are worth investigating. That framing positions Inherent as something structurally different from the generation of AI science tools that preceded it.

The team behind Faraday

Co-founders Tantum Collins and Edward Hughes worked together on cooperative AI research at DeepMind. Louis Kirsch, a third co-founder, also came from DeepMind. Kaloyan Aleksiev brings experience from Reka AI and Microsoft. Collins is unusual among AI lab founders for holding a policy background: he worked on AI matters at the Biden White House before shifting to company-building. Matt Clifford, the UK government's former AI tsar and co-founder of Entrepreneurs First, has joined as an adviser, lending the venture credibility across both technical and policy circles.

What Faraday actually does

Faraday combines human researchers with AI agents that iterate on scientific problems over time, improving their approach with each cycle. Inherent describes the paradigm as "AI-native science," a departure the company claims will look and feel different from the scientific method as practiced over the past four centuries. That is a large claim. How the system determines which questions merit pursuit, and by what criteria it evaluates novelty or tractability, was not specified in the launch materials reviewed by The Next Web.

Danny Rimer of Index Ventures articulated the investment thesis through analogy, in remarks reported by The Next Web. Open-ended curiosity produced penicillin, the microwave, and the GPU; standard question-answering AI cannot replicate that kind of exploratory process, he argued. The gap Inherent is targeting sits upstream of computation: formulating the question before any experiment begins.

Why practitioners should pay attention

The distinction between question-answering and question-generation carries real weight in domains like artificial intelligence in medicine, where identifying the right protein target or drug interaction to investigate can matter as much as the eventual result. Tools like AlphaFold demonstrated that AI excels at answering well-posed structural biology questions. The harder problem is earlier in the pipeline: deciding which questions are well-posed to begin with.

Collins's White House policy background may prove more relevant than it first appears. Scientific priority-setting is political as much as intellectual, shaped by funding agencies, regulatory constraints, and institutional incentives. An AI system that influences which research directions get pursued will operate at exactly that intersection, a terrain that purely technical founding teams rarely navigate well.

The funding picture

A $50 million seed is substantial by any measure, but it reflects where early-stage AI infrastructure bets are landing in 2026. NVentures' participation signals that Nvidia sees potential hardware demand in whatever compute-intensive process Faraday runs at scale. Inherent has not disclosed revenue, a product launch timeline, or which scientific domains Faraday will target first, per The Next Web.

Automated hypothesis generation is not a new idea. Researchers explored it in symbolic AI systems in the 1980s, and meta-learning revived the concept in the neural era. What has changed is the combination of large language models capable of cross-domain reasoning with reinforcement learning methods that can improve agent behavior on open-ended tasks. Whether those ingredients are sufficient to replicate the kind of serendipity behind landmark discoveries is a question the artificial intelligence review literature has not settled. For ML engineers evaluating tools in this space, the absence of benchmark disclosures at launch is a meaningful signal to watch.

Inherent has assembled a credible team and articulated a genuine problem. The real test arrives when Faraday produces results: does it surface research directions that scientists would not have reached on their own, or does it optimize within the distribution of questions that are already being asked? That is the only benchmark that matters here.

FAQ

Q: What is Inherent AI's Faraday platform?
Faraday pairs human researchers with self-improving AI agents designed to identify which scientific questions are worth pursuing, rather than answering queries that researchers have already formulated.

Q: Who founded Inherent AI?
The team includes Tantum Collins and Edward Hughes, who collaborated on cooperative AI research at DeepMind, along with Louis Kirsch (also from DeepMind) and Kaloyan Aleksiev from Reka AI and Microsoft.

Q: How does AI determine which scientific questions to ask?
Inherent has not published technical specifics. The company describes an iterative self-improvement process where AI agents work on hard scientific problems alongside human researchers, refining their approach over successive cycles.

Q: What makes Inherent different from existing AI science tools?
Most AI science tools, including large-scale protein-folding systems, answer well-defined questions. Inherent's premise is that the bottleneck in scientific discovery is upstream: question formulation, not computation speed.