Inherent Raises $50M to Build AI That Selects Research Questions

A $50 million bet on a deceptively simple idea: the real bottleneck in science is not compute or data, it is knowing what to investigate.

Inherent, a London-based AI lab, emerged from stealth on Wednesday with one of Europe's largest seed rounds of 2026. Index Ventures and Radical Ventures co-led the raise; Nvidia's venture arm NVentures joined alongside Ex/Ante, Metaplanet, Macroscopic Ventures, and Mythos Ventures. The Next Web ranked it among the continent's biggest stealth-to-launch rounds this year.

Three of the four co-founders came through DeepMind. Tantum Collins and Edward Hughes previously collaborated on cooperative AI research there; Louis Kirsch also spent time at the lab. Kaloyan Aleksiev rounds out the team, coming from Reka AI and Microsoft. Collins carries a credential rare among deep-tech founders: he worked on AI policy at the Biden White House before pivoting to company building.

The question Faraday is trying to answer

Inherent's product is called Faraday, named after the Victorian physicist who had no formal university training but produced foundational work in electromagnetism through directed curiosity. The name is deliberate. The company's core argument is that modern artificial intelligence is structurally optimized for answering questions, not generating them.

"Most AI is built to answer questions," said Danny Rimer of Index Ventures. "What it can't do yet is figure out which questions are worth asking - the open-ended curiosity that produced penicillin, the microwave, the GPU." Faraday matches human researchers with AI agents built to improve iteratively on hard scientific problems. Inherent calls this paradigm "AI-native science" and frames it as a fundamental departure from how scientists have worked for the past four centuries.

That phrase is new; the underlying tension is not. Research attention has always been allocated inefficiently: scientists pursue fundable questions, peer review rewards novelty-within-paradigm, and high-value neglected problems accumulate. A platform that reliably surfaces those problems could have compounding benefits across every discipline. Whether Faraday can actually deliver is exactly the bet, and the company has not disclosed which scientific domains it will target first.

Domain specificity matters more than the pitch lets on. A self-improving agent that works in protein structure prediction will fail in qualitatively different ways from one aimed at drug candidate prioritization or climate modeling. The pitch is domain-agnostic; the hard work is not.

The policy angle

Collins's White House background is not incidental. Automating research prioritization is partly a technical problem and mostly an institutional one. Universities, funding agencies, and journals move slowly and resist externally generated research agendas. A tool that surfaces the right questions is useless if the researchers who receive them cannot act on them.

Matt Clifford, co-founder of Entrepreneurs First and the UK government's former AI adviser, has joined as an adviser to Inherent. His presence suggests the company is thinking seriously about how its artificial intelligence platform integrates with existing research governance structures, a question no lab has cleanly resolved. CNET noted this week that the broader AI market is crowded with companies racing to bring genuinely new capabilities to research and enterprise contexts.

The competitive artificial intelligence landscape for science has expanded rapidly. AI Release Tracker documents over 160 frontier models released since late 2022, many now targeting scientific applications. Inherent is not competing on benchmark performance. Its pitch is one level up: infrastructure for deciding which problems are worth running those models on at all. That is a harder market to win because success criteria are diffuse and long-tailed, but it is also far less crowded.

Fifty million dollars at seed is at the high end even in a year when large pre-revenue rounds have become routine. Inherent has no disclosed revenue, no published research, and no announced institutional partners. Index and Radical are betting on team pedigree and thesis clarity, a coherent wager given the team's cooperative AI and reinforcement learning backgrounds, but one that leaves substantial distance between the pitch and a working product.

What comes next

Inherent's forthcoming disclosures will determine whether Faraday represents a genuinely novel infrastructure layer for science or an expensive reframing of AI-assisted literature review. The company has not announced research partnerships, a product timeline, or benchmarks of any kind.

The harder question behind all of this: can you automate research selection without encoding the system builders' assumptions about what counts as valuable? Faraday may prove as capable as its founders claim at surfacing high-impact problems. The risk is that it learns to optimize for the wrong thing, and no one notices until the science is already downstream of the mistake.

---

FAQ

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

Q: Who founded Inherent?
A: Tantum Collins, Edward Hughes, and Louis Kirsch all came from DeepMind, where Collins and Hughes worked on cooperative AI research. Kaloyan Aleksiev joined from Reka AI and Microsoft. Collins also worked on AI policy at the Biden White House before co-founding the company.

Q: Who are Inherent's investors?
A: Index Ventures and Radical Ventures co-led the $50 million seed round. Nvidia's NVentures, Ex/Ante, Metaplanet, Macroscopic Ventures, and Mythos Ventures also participated.

Q: How does Faraday differ from tools like AlphaFold?
A: AlphaFold solves a well-defined problem in structural biology. Inherent's pitch is that Faraday operates one level up, identifying which problems are worth solving before any domain-specific model is applied - what the company calls "AI-native science."