Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research

Inherent, a UK-based AI startup founded by former DeepMind researchers, has launched an AI agent called Faraday that reportedly surpasses offerings from Anthropic and OpenAI in its ability to accurately reproduce the findings and methodologies of published scientific research papers.

A British artificial intelligence company called Inherent, established by veterans of Google's DeepMind research lab, has unveiled an AI agent named Faraday. The tool is specifically designed to replicate the experiments and conclusions found in academic scientific papers — a notoriously difficult task that requires careful reasoning and methodological precision.

According to Inherent, Faraday has outperformed comparable AI systems from industry giants Anthropic and OpenAI on this benchmark. The ability to reliably reproduce scientific research is considered an important capability because it could accelerate the pace of discovery, help identify flawed or fraudulent studies, and serve as a foundation for AI systems that can eventually conduct original research autonomously.

Inherent, a United Kingdom-based AI laboratory staffed by alumni from Google's prestigious DeepMind research division, has publicly launched an AI agent it calls Faraday. The system is purpose-built around one of the more technically demanding challenges in science: taking a published research paper and faithfully replicating its experiments, data analysis, and conclusions from scratch.

The company claims that in head-to-head evaluations, Faraday outperformed AI agents developed by Anthropic and OpenAI on this specific research-replication task. While independent third-party verification of such claims is important context, the benchmark itself is meaningful — replicating scientific work requires an agent to understand complex methodologies, execute multi-step reasoning, and produce consistent, accurate outputs rather than plausible-sounding but incorrect ones.

Why it matters: The so-called 'replication crisis' has plagued academic science for years, with many published studies proving impossible to reproduce, raising questions about their validity. An AI system capable of rapidly and reliably replicating experiments could serve as a powerful quality-control layer for the scientific community. Beyond verification, such a capability is widely seen as a prerequisite for AI systems that can eventually move from reproducing existing knowledge to generating genuinely novel scientific discoveries.

The founding team's DeepMind pedigree also adds credibility to the effort. DeepMind itself has a strong track record at the intersection of AI and science, most notably through AlphaFold, which revolutionized protein structure prediction. Inherent appears to be positioning Faraday as a collaborative 'teammate' for researchers rather than a replacement, framing the tool as something that amplifies human scientific productivity.

For the broader AI industry, this launch signals growing competition in the 'AI for science' niche, where well-funded startups with deep research talent are increasingly challenging the dominant general-purpose labs. How Faraday performs under independent scrutiny will be the real test of whether its early benchmark results translate into practical scientific value.