Inherent is a London-based artificial-intelligence laboratory that emerged from stealth on May 29, 2026 with a $50 million seed round and a stated aim of building AI systems for scientific discovery. It was founded by three former Google DeepMind researchers — Tantum Collins, Edward Hughes and Louis Kirsch — together with Kaloyan Aleksiev, previously of Reka AI and Microsoft (Source: tech.eu). Its first published system, Faraday, is a 27-billion-parameter agent trained to replicate figures from research papers (Source: inherentlabs.ai).
Infobox
| Field | Value |
|---|---|
| Headquarters | London, United Kingdom |
| Founded | Emerged from stealth May 29, 2026 |
| Founders | Tantum Collins, Edward Hughes, Louis Kirsch, Kaloyan Aleksiev |
| Seed funding | $50 million (£37.2 million), co-led by Index Ventures and Radical Ventures |
| Adviser | Matt Clifford |
| Principal system | Faraday |
| Web | inherentlabs.ai |
Snapshot
Funding
| Date | Amount | Round | Investors | Source |
|---|---|---|---|---|
| 2026-05-29 | $50M | Seed | Index Ventures and Radical Ventures (co-leads) | (Source: tech.eu) |
Reporting on the round is not fully consistent on lead-investor structure: tech.eu and The Next Web describe Index Ventures and Radical Ventures as co-leads, while Dealroom describes the round as led by Index Ventures with Radical Ventures participating (Source: tech.eu; thenextweb.com; dealroom.co). The dollar figure is consistent across all accounts.
Founders and people
Tantum Collins, Edward Hughes and Louis Kirsch each worked at Google DeepMind before founding the company; Collins previously served in the United States White House (Source: tech.eu). Kaloyan Aleksiev previously worked at Reka AI and Microsoft (Source: dealroom.co). Matt Clifford, co-founder of Entrepreneurs First and a former United Kingdom government AI adviser, is an adviser to the company (Source: tech.eu).
The August 2026 Faraday paper carries eleven authors: Damon Falck and Samer Sabri as equal first authors, then Anja Surina, Thom Foster, Anya Sims, Sam Devlin, Dylan Rogers, Tantum Collins, Kaloyan Aleksiev as infrastructure lead, and Louis Kirsch and Edward Hughes as equal last authors (Training AI Scientists to Replicate Research).
Stated approach
The company describes its object as writing "the playbook for AI-native science" and reimagining the scientific method rather than applying AI to existing methods. Danny Rimer of Index Ventures, quoted in the launch coverage, characterised Faraday as "a system designed to help humans and self-improving AI work together on genuine scientific discovery — not AI plugged into the same methods we've used for 400 years, but a reimagining of the scientific method from first principles," and separately described what current systems cannot yet do as "figure out which questions are worth asking" (Source: tech.eu; dealroom.co).
In its own research writing the company frames replication as a route to open-ended research: because papers report what worked and not the negative results behind it, an agent asked to reproduce a figure must recover the intermediate exploration, which the company argues requires the same hypothesis-driven search as original work. It states that replication tasks form the basis of "a curriculum of underspecified tasks" in which more of the paper can be withheld, resource constraints tightened, and papers themselves imagined (Source: inherentlabs.ai).
The company states that it is investigating how its methods "might advance scalable oversight and ameliorate reward hacking," and describes Faraday's use inside the company as "enabling us to discover new knowledge while firmly keeping humans in the loop" (Source: inherentlabs.ai). No safety framework, responsible-scaling policy or model-release policy had been published as of August 16, 2026.
Research and systems
The company's first published research, "Training AI Scientists to Replicate Research," was posted to arXiv on August 13, 2026 and announced on the company's site on August 14 (Training AI Scientists to Replicate Research; Source: inherentlabs.ai). It introduces two artefacts: Replica, a task space of 310 paper-replication tasks drawn from 100 machine-learning and AI-for-science papers, and Faraday, a 27-billion-parameter agent post-trained on that task space. Both are described on Faraday. Neither the model weights nor the task space was released, and the paper reports the comparison against Claude Opus 4.8 and GPT-5.5 on the company's own rubric-based judge.
The paper's ethics statement discloses that some papers in the Replica corpus were written by its own authors or by people they know personally, and that some come from institutions whose commercial models the research used — Anthropic and OpenAI models serve as rubric generator, judge and coding-agent tool throughout. The authors state that the same automated pipeline was applied to every item in the corpus with no altered scoring treatment (Training AI Scientists to Replicate Research).
At the May 2026 launch, "Faraday" was reported as the name of the platform the company was building — an AI system "that allows humans and self-improving AI to work together" on hard scientific problems (Source: tech.eu). The August 2026 paper uses the name for a specific 27B model. Whether the two are the same artefact under a shared name, or a platform and a model within it, is not stated in either source. The name refers in both cases to the nineteenth-century experimentalist Michael Faraday, whose replication of earlier electromagnetism results the company's research post recounts as its framing example (Source: inherentlabs.ai).
Naming
Sources published by and about the company use the single word "Inherent"; the corporate web presence is at inherentlabs.ai, job listings are posted under the handle "inherent", and the August 2026 paper gives the authors' affiliation as "Inherent" (Source: inherentlabs.ai; tech.eu). The longer form "Inherent Laboratories" appears in the paper's own copyright line — "© 2026 Inherent Laboratories" — and in its correspondence address, faraday@inherentlaboratories.com, which are the only primary uses recorded (Training AI Scientists to Replicate Research). Secondary summaries that render the company as "Inherent Laboratories" are therefore consistent with the corporate name even though every editorial use by the company is the short form.
Relationships
- related: Faraday — the company's first published system
- related: Google DeepMind — the prior employer of three of the four founders
- related: AI for Science — the application area the company is built around
- related: Scalable Oversight, Reward Hacking — safety questions the company states it is investigating through this work
- related: Recursive Self-Improvement (RSI) — the "self-improving AI" framing used in the company's launch coverage
- related: Training AI Scientists to Replicate Research — the company's first published research
Open questions
- Whether the "Faraday" platform announced in May 2026 and the 27B Faraday model published in August 2026 are the same artefact is not stated by any source.
- The company has published no safety framework, model-release policy or evaluation commitments.
- No independent party had reproduced the Replica results as of August 16, 2026.
Sources
- Inherent / "Training AI Scientists to Replicate Research" (2026-08-14) — inherentlabs.ai
- Training AI Scientists to Replicate Research — Falck et al., "Training AI Scientists to Replicate Research," arXiv:2608.13331 (2026-08-13)
- tech.eu / "London-based AI lab Inherent emerges from stealth with $50m raise" (2026-05-29)
- Dealroom / "Inherent emerges from stealth with $50M seed to build AI-native science"
- The Next Web / "Ex-DeepMind researchers raised $50M to build AI that figures out what to ask"
- UKTN / "London-based AI lab Inherent emerges from stealth with £40m raise" (2026-05-29)