동향

자율실험실과 AI 소재 발견 분야의 주목할 만한 소식——인재 이동, 신규 연구소, 자금, 정책.

2026-10

마일스톤 ¶

Danaher announces its first AI-powered autonomous lab, for antibody and affinity-reagent discovery at Abcam

Danaher announced plans to launch its first AI-powered autonomous lab, based at Abcam (a Danaher company) and expected to operate at scale in early 2027. The lab uses AI to design new affinity reagents, including antibodies, then builds and tests each one and feeds every result back into the model in a continuous design-make-test-learn loop, with scientists making the key decisions. It combines technologies from several Danaher operating companies (Beckman Coulter Life Sciences, Cytiva, Genedata, Integrated DNA Technologies and Molecular Devices) in one connected workflow; device orchestration and robotic automation were developed with Automata. Danaher targets up to eight times faster progress from idea to validated reagent and up to a tenfold increase in annual reagent generation, from tens to hundreds per year. The company describes the lab as the first step of a broader programme to build smart instruments that are easier to control programmatically, produce AI-ready data and include agentic systems, connected into integrated workflows and autonomous labs.

Danaher press release (7 Oct 2026) ↗ → 자세히 보기

2026-09

마일스톤 ¶

Roche outlines its move towards autonomous AI labs for drug R&D, with Medra and Anthropic as collaborators

At its Pharma Day investor event on 28 September 2026, Roche outlined plans to move towards autonomous AI-driven labs to speed up drug discovery. Aviv Regev, head of Genentech Research and Early Development, said the company is committed to 'AI independence' in R&D and has recently started building autonomous labs. An Autonomous Lab Accelerator is meant to strengthen Roche's lab-in-the-loop model, in which AI models make predictions, lab experiments test them and the results are fed back into the models. Roche presented a six-level scale for drug R&D labs, from fully manual (L0) through automated instruments, connected workflows, AI-assisted workflows and domain autonomy to lab-wide autonomy (L5), with the human role shifting from operator through assistant and approver to auditor as autonomy rises. The deck names Medra (physical AI) and Anthropic, including its Model Hardware Standard, as autonomous-lab collaborators, next to NVIDIA for a hybrid-cloud AI factory of more than 3,500 Blackwell GPUs and Recursion for data generation. Roche reported that 40% of pipeline decisions from Q4 2025 to Q2 2026 had a tracked AI or computational contribution, and expects its Target Nexus AI tool to contribute to 80% of research portfolio decisions by the end of 2026.

Reuters (28 Sep 2026) ↗Roche Pharma Day 2026 presentation (slides 71, 94, 100) ↗Roche Pharma Day 2026 ↗ → 자세히 보기

펀딩 ¶

EU opens a €29M 'Automated Scientific Discovery' call: the first Horizon Europe topic dedicated to self-driving labs (RAISE pilot)

The European Commission opened HORIZON-RAISE-2027-01-01 'Automated Scientific Discovery (RAISE pilot)', a Research and Innovation Action with €29M for about three projects of €10M each, single-stage deadline 2 February 2027. The topic funds labs that already have an advanced level of automation and digitalisation to build the intelligence layer that lets their instruments plan, run and analyse experiments semi- or fully autonomously in closed loop, with human oversight for accuracy, safety and ethics, ideally networked with other labs and without a complete redesign of existing equipment. Proposals must include data management for collecting, processing and sharing experimental data across labs; research targets include agentic AI connected to instruments and robotics, and resource optimisation that cuts energy, reagent and material use. The call sits inside RAISE, the Resource for AI Science in Europe, the EU's virtual institute for AI in science launched in Copenhagen in November 2025, with €107M from Horizon Europe for its 2026–27 pilot.

EU Funding & Tenders portal (topic HORIZON-RAISE-2027-01-01) ↗European Commission: RAISE ↗Science|Business ↗

2026-08

마일스톤 ¶

Anthropic previews the Model Hardware Standard (MHS) — one interface for AI agents to operate laboratory instruments

Anthropic opened a research preview of the Model Hardware Standard, a shared, model-agnostic specification through which an AI agent reads sensors, drives actuators and operates instruments such as microscopes, liquid handlers and robotic arms, several in parallel. It began as a collaboration with HHMI Janelia Research Campus and launches with a first cohort of laboratories and manufacturers: Tecan is adding MHS to its Fluent liquid handlers, Automata to its LINQ platform, QIAGEN is prototyping it on QIAsymphony Connect, Universal Robots and Doosan Robotics on their arms, AWS through Strands Robots, MBF Bioscience for ScanImage microscopes, with Danaher exploring. Reported tests include a BCA protein assay at Genentech coordinated across a liquid handler, an arm and a plate reader (human experts stepped in when the agent misread bubble errors as software faults), and laser stabilisation on QuEra's quantum computers from 58% to 99.3%. Anthropic says it will open-source the specification after the preview's safety evaluations. For self-driving labs this is the agent-to-instrument layer that vendor APIs and SiLA 2 have left fragmented, arriving from a model vendor rather than an instrument maker; three weeks later Anthropic also announced an AI-run biology laboratory of its own.

Anthropic announcement ↗CNBC ↗R&D World ↗SiliconANGLE (biology lab, 18 Sep) ↗ → 자세히 보기

마일스톤 ¶

Chem hosts the field's first structured debate on 'reasoning SDLs' — Abolhasani Catalyst + five Reaction responses

Chem's August 2026 issue (Vol. 12/8) pairs Milad Abolhasani's Catalyst commentary 'When self-driving labs begin to reason' — arguing the SDL bottleneck has moved from executing experiments to deciding what science to do next — with five invited Reaction responses from leading groups: Chen/Xie/Jiang (USTC) call for experimentally grounded chemical world models; Boser/Glorius (Münster) bring synthetic-chemistry adoption realism; Fan/Liu (UCLA) demand agentic AI that explains as well as optimizes; Aspuru-Guzik/Bernales (Toronto) argue for trust infrastructure 'beyond fragile eloquence'; and Sin/Segura/Winkler/Schwaller (EPFL LIAC + Roche) frame reasoning labs as reproducible physical infrastructure. Notably, every response converges on verification and infrastructure rather than raw capability — the same month the anchor author's Nature Reviews Chemistry review recast SDLs as shared scientific infrastructure.

Abolhasani Catalyst (Chem) ↗Aspuru-Guzik & Bernales Reaction ↗Schwaller et al. Reaction ↗ → 자세히 보기

펀딩 ¶

UT Knoxville wins $20M NSF award for ATHENA — autonomous microscopy + synthesis cloud testbed led by Sergei Kalinin

The University of Tennessee, Knoxville secured a $20M NSF award to build ATHENA (Advanced Testbed for High-throughput Experimentation in Nano- and Atomic Science), a node in NSF's Programmable Cloud Laboratories initiative. Led by Sergei Kalinin — a defining figure in autonomous microscopy — with co-PIs Gerd Duscher, Mahshid Ahmadi and Hairong Qi and twelve UT faculty, partnering Northwestern and Johns Hopkins, ATHENA combines autonomous microscopy with automated materials synthesis into remotely accessible, cloud-operated self-driving laboratories built on open software standards, targeting a 10–30× increase in experimental throughput. A companion $5M Johns Hopkins-led award on an AI for Materials Design Network includes UT's Michela Taufer.

UT Knoxville (Tickle College of Engineering) ↗NSF PCL Test Bed (program) ↗ → 자세히 보기

마일스톤 ¶

Korea launches a national Autonomous-Laboratory Consortium — 80 institutions behind the K-Moonshot lab build-out

MSIT convened the Autonomous Laboratory Industry–Academia–Research Consortium in Seoul: ~50 companies (Samsung SAIT, SK Hynix, LG Chem, Hyundai Motor, Park Systems) and ~30 universities and institutes (KIST, KIMM, KIMS, KRIBB), chaired by SNU's Yousung Jung with the National AI Science Research Center and KBSI as secretariat. The $36.5M two-year primary programme builds one general-purpose autonomous lab, five specialized labs and four autonomous experiment centers (secondary batteries, hydrogen, aerospace/mobility, organic semiconductors/displays), and standardizes equipment, protocols and data formats — feeding national AI training datasets under the K-Moonshot initiative. Alongside MOTIE's 500-SDL commitment, Korea now has both an industrial target and an institutional coordination body.

HelloDD (헬로디디) ↗Herald Business (헤럴드경제) ↗Science and Medicine Group (EN summary) ↗ → 자세히 보기

2026-07

펀딩 ¶

NSF commits ~$400M to a national network of AI-driven 'programmable cloud laboratories'

The US National Science Foundation launched the 'NSF Test Bed: Toward a Network of Programmable Cloud Laboratories (NSF PCL Test Bed)', awarding $380M to 20 teams over four years, with upwards of $20M in matching philanthropic funding from the Astera Institute (~$400M total). The teams will build a nationwide, remotely accessible network of AI-enabled automated laboratories aiming for 'a virtuous cycle of automated hypothesis generation, autonomous experimentation and generation and interpretation of large volumes of high-quality experimental data', across disciplines from biology and chemistry to soft materials, 2D materials, metals, material characterisation and electronics; several PCL Nodes include participation by the US DOE National Labs. One example node: Penn State's $20M 'LATTICE' (Layered and Thin-film Technologies with Intelligent Cloud Experimentation), led by Distinguished Professor Joan Redwing with Argonne National Laboratory, for AI-enabled thin-film materials discovery, characterisation and optimisation.

NSF ↗NSF PCL Test Bed (program) ↗Penn State (LATTICE node) ↗ → 자세히 보기

마일스톤 ¶

Cusp AI launches an 'AI Materials Foundry' and raises a $450M Series B at a roughly €2.2B valuation

Cusp AI, the AI-materials-discovery startup co-founded by Max Welling (Amsterdam-based) and Chad Edwards, has launched the 'AI Materials Foundry': a global network of 45+ founding partners (NVIDIA, Meta, Samsung, Hyundai Motor Group, Henkel, Applied Materials, Tokyo Electron, Lam Research) pooling compute and scientific resources to accelerate materials discovery for chipmaking and other industries. The launch came with a $450M (€393M) Series B at a roughly €2.2B ($2.6B) valuation, led by Kleiner Perkins and NEA with Jeff Bezos' Bezos Expeditions participating. A privately funded, Welling-anchored push on exactly the AI-driven materials-discovery frontier DiscoveryLabNL is building national public infrastructure for.

Cusp AI (Medium) ↗EU-Startups ↗Bloomberg ↗Silicon Republic ↗ → 자세히 보기

인재 ¶

Nobel laureate Omar Yaghi moves to Tsinghua to lead an AI materials-discovery institute

Omar Yaghi (2025 Chemistry Nobel; the pioneer of metal-organic frameworks) has left UC Berkeley for a full-time chair at Tsinghua University in Beijing, where he will lead a new AI-assisted materials-discovery institute. A high-profile US-to-China move on exactly the frontier DiscoveryLabNL is building for.

Nature ↗NYT ↗Tsinghua ↗ → 자세히 보기

마일스톤 ¶

MaterialsCommons launches: a €28M EU federated digital infrastructure for materials R&D, with a framework for self-driving labs

MaterialsCommons ('Federated Digital Infrastructure for European Advanced Materials Research and Development', Horizon Europe Innovation Action, grant 101293545, about €28M) started on 1 June 2026 and runs to May 2030. It is coordinated by the Fraunhofer-Gesellschaft (Peter Gumbsch, Fraunhofer IWM), with EMMC and DTU as vice-coordinators, and brings together 26 research institutions and more than 30 industrial partners across 14 countries. The project connects existing European digital resources into a single entry point for materials data, workflows and tools, built on three core technologies: semantic interoperability, interoperable workflows for experiments and simulations, and federated dataspaces. Its targets include a TRL6 prototype handling 10,000 queries a day and integrating over 90% of widely used data sources, fully automated workflows, and a framework for self-driving labs that combines AI and machine learning to raise experimental throughput and shorten design-to-validation cycles. In its use cases the consortium aims to accelerate materials and component development by up to tenfold. HaDEA presented the project on 3 July 2026.

HaDEA (Q&A with Peter Gumbsch) ↗CORDIS (project 101293545) ↗MaterialsCommons ↗ → 자세히 보기