Anthropic AI for Science 2026
John Jumper · Claude Mythos 5 · Drug discovery 10× speed

TL;DR: 30 июня 2026 Anthropic проводит The Briefing: AI for Science — крупнейший life sciences moment после Claude Code. Nobel laureate John Jumper (AlphaFold co-creator) 19 июня ушёл из Google DeepMind в Anthropic. Claude Mythos 5 drug design ~10× faster, strong compounds на 9/14 protein targets zero human help. Novo Nordisk CSR writing −90%. CEOs Novartis, BMS, Genentech on stage. Guide covers event lineup, Jumper AlphaFold legacy, 18-month Claude for Life Sciences rollout, Mythos 5 benchmarks, Coefficient Bio acquisition, export control status + life sciences FAQ — six context modules pre-livestream.

01

6 модулей контекста перед Anthropic AI for Science

Это не routine model drop. Anthropic объявляет life sciences следующей major vertical после Claude Code. Шесть hard facts отделяют signal от hype до livestream 30 июня.

  1. 01

    Jumper — headline, product — Claude for Life Sciences: Nobel move тянет PR, но real bet — enterprise stack с Benchling, PubMed, 10x Genomics, Medidata; launch Oct 2025, production у Novo Nordisk, Sanofi, AbbVie.

  2. 02

    Pharma CEOs on stage = enterprise lock-in: Vas Narasimhan (Novartis CEO + Anthropic board), Chris Boerner (BMS), Aviv Regev (Genentech R&D) — это production deployments, не pilot sandbox.

  3. 03

    Mythos 5 benchmarks real, export-controlled: 10× drug design speed, 9/14 autonomous hits — global shutdown 12 июня. На 30 июня только ~100 US critical-infrastructure orgs с partial restore.

  4. 04

    Coefficient Bio — скрытый strategic move: ~$400M all-stock acquisition в апреле; sub-10 Genentech Prescient Design alumni — bridge от LLM assistant к drug discovery engine.

  5. 05

    18 месяцев deliberate buildup: Claude for Life Sciences (Oct 2025) → Allen Institute partnerships → Karpathy pre-training (May 2026) → Fable/Mythos 5 (Jun 9) — сегодня public capstone.

  6. 06

    Developer execution layer всё ещё bottleneck: Claude Code + MCP для toxicity prediction (Axiom) требует stable long-running Agent sessions. 16GB laptop swap на multi-hour genomics pipeline — см. тарифы аренды cloud Mac как execution substrate.

02

Что такое Anthropic AI for Science event?

The Briefing: AI for Science — 30 июня 2026, 10:00 AM PST (20:00 MSK) — hybrid event из San Francisco + global livestream.

ParameterValue
Event nameThe Briefing: AI for Science
Date & time30 June 2026 · 10:00 AM PST
FormatSan Francisco in-person + global livestream
HostAnthropic (Claude parent co.)
AgendaLife sciences vision, Claude for Life Sciences demos, top pharma customer spotlights

Speaker roster

SpeakerRole
Vas NarasimhanCEO Novartis; Anthropic board member
Chris Boerner, PhDCEO Bristol Myers Squibb
Aviv RegevEVP & Head of R&D, Genentech
Lotte Bjerre KnudsenProfessor, DMSc; ex-Novo Nordisk Chief Scientific Adviser
Eric Kauderer-AbramsHead of Life Sciences, Anthropic
Jonah CoolHead of Life Sciences Partnerships, Anthropic
Matthew HerperSenior Writer, STAT News (moderator)

Когда CEO Novartis и BMS делят stage с AI company — это не marketing stunt, а vector реального pharma R&D.

Livestream watchlist

  • Jumper появится? Role undisclosed; даже brief cameo = major signal.
  • Mythos 5 biology access expansion для vetted US orgs
  • New Claude for Life Sciences connectors / Agent Skills
  • Fable 5 restoration timeline — sources: resolution «as soon as this week»
  • International access path для non-US researchers под export lock
03

John Jumper: кто это и почему move в Anthropic — game changer

Человек, закрывший 50-летнюю biology problem

John Michael Jumper, b. 1985 Little Rock AR: Vanderbilt (BS math/physics 2007), Cambridge (MPhil physics, Marshall 2008), UChicago (PhD theoretical chemistry 2017).

+6 months post-PhD → Google DeepMind secret project: protein folding 3D structure prediction — open problem с 1970-х.

CASP14 2020: AlphaFold 2 hits experimental-grade accuracy — hours vs months.

AlphaFold impactScale
Structures predicted214M+ across ~1M species
Global adoption2M+ users, 190+ countries
ApplicationsDrug discovery, cancer research, molecular biology

2024: Jumper + Demis Hassabis Nobel Prize in Chemistry (+ David Baker UW). Age 39 — youngest chemistry laureate in 70+ years.

Почему DeepMind на peak?

19 June 2026, X post:

“After nearly nine years, I have decided to leave Google DeepMind and join Anthropic.”

Hassabis public reply: “We changed the world with AlphaFold” — AI path for science and medicine.

Timing: 11 days pre-event — deliberate. Specific role TBD. Bloomberg Jun 24: co-authors Adler, Pritzel may follow — unconfirmed.

Strategic fit: Apr 2026 Anthropic acquired Coefficient Bio (~$400M), founders from Genentech computational drug discovery. Jumper sits at intersection foundational biology AI ↔ applied drug design.

04

18-month rollout: Claude for Life Sciences stack breakdown

Сегодня — public capstone systematic rollout, не day-one pivot.

DateMilestone
October 2025Claude for Life Sciences launch; anchor: Novo Nordisk, Sanofi, AbbVie, AstraZeneca, Genmab
February 2026Allen Institute + HHMI Janelia research partnerships
April 2026Coefficient Bio ~$400M (ex-Genentech comp bio team)
May 19, 2026Andrej Karpathy joins pre-training team
June 9, 2026Claude Fable 5 + Mythos 5 science capabilities launch
June 12, 2026US export control: Fable 5/Mythos 5 global shutdown
June 19, 2026John Jumper DeepMind → Anthropic
June 26, 2026Mythos 5 partial restore ~100 US orgs
June 30, 2026The Briefing: AI for Science

Claude for Life Sciences — что внутри

Oct 2025 launch: vertical on Claude Enterprise — MCP connectors + Agent Skills wired в existing scientist toolchain.

PlatformClaude function
BenchlingSOPs, informed consent, protocols from ELN/LIMS
10x GenomicsSingle-cell + spatial transcriptomics analysis
PubMedLiterature search/retrieve/synthesize + citations
bioRxiv / medRxivPreprint retrieval + analysis
Open TargetsTherapeutic target ID + prioritization
MedidataClinical trial enrollment + site performance
ClinicalTrials.govRegistry queries + competitive landscape
Wiley Scholar GatewayFull-text journal access
BioRenderScientific image processing

R&D stages covered

  • Early discovery: Literature review, hypothesis gen, target ID, experimental design
  • Preclinical: Genomics, scRNA-seq QC, toxicity prediction
  • Clinical trials: Protocol drafting (FDA/NIH), enrollment monitoring, risk alerts
  • Regulatory: Doc drafting, gap analysis, FDA query responses, CTD automation
05

Claude Mythos 5: benchmarks, pharma cases, Coefficient Bio, export control

Drug design: 10× speed, fully autonomous pipeline

Internal Anthropic testing: Mythos 5 accelerates key drug design steps ~10×. Zero human assistance study — full scientific workflow end-to-end:

  1. 1

    Binding site identification

  2. 2

    Protein design tool selection + execution

  3. 3

    Autonomous failure recovery

  4. 4

    Viable drug candidate output under investigation

Result: 9/14 protein targets (64%) strong candidates — immune checkpoints, growth-factor signaling, neurodegeneration, muscle disease, complex structural targets.

Beating domain specialists head-to-head

  • AAV capsid prediction (Dyno Therapeutics dataset): Mythos 5 > dedicated protein LMs
  • Hypothesis generation: Blind review ~80% preference vs prior Opus; one E. coli antimicrobial hypothesis lab-validated
  • Autonomous genomics (1 week unsupervised): scRNA data from 138 animal species; custom ML model 100× smaller than recent Science pub, better perf

Production pharma deployments

Novo Nordisk (Ozempic mfg) built NovoScribe on Amazon Bedrock + Claude + RAG:

“Claude has helped us cut writing times on CSRs by 90% so we can get documentation directly into human hands for review and approval.” — Waheed Jowiya, Digitalization Strategy Director, Novo Nordisk

NovoScribe expanding → device protocols, patient materials, full CTD automation.

Confirmed deployments: Sanofi, AbbVie, AstraZeneca, Genmab, BMS, Komodo Health, Axiom (toxicity via Claude Code + MCP DB servers).

Coefficient Bio: highest-signal acquisition

Apr 2026: Coefficient Bio ~$400M all-stock, <10 FTE — Samuel Stanton, Nathan C. Frey from Genentech Prescient Design, «ASI for Science» thesis. Dimension investor: 38,513% IRR reported. Team under Eric Kauderer-Abrams — protein design + biomolecule modeling bridge to true AI drug discovery engine.

Industry context: why now

  • Traditional drug dev: 12–15 years, $2.6B+/approved drug (2024); ~10% clinical candidates reach market
  • AI compresses target ID months→hours, compound design by orders of magnitude, regulatory doc drafting by multiples
  • Anthropic edge: Constitutional AI for compliance-sensitive pharma, vertical integration (platform + Coefficient Bio + Jumper), locked top-tier customers

Export control: Mythos 5 access matrix

Mythos 5 drives these benchmarks — access severely restricted:

DateExport control status
June 12, 2026US Commerce bans foreign nationals from Fable 5/Mythos 5; global shutdown <90 min
June 26, 2026Partial restore: ~100 US critical-infrastructure orgs
June 30, 2026Fable 5 still offline; negotiations ongoing; general access not restored

Multinational pharma teams — including non-US staff at today’s stage companies — face real ops uncertainty. See export control survival guide.

warning

Can Jumper replicate AlphaFold at Anthropic? Genuinely uncertain. AlphaFold needed DeepMind infra, institutional partnerships, CASP benchmark. Anthropic = commercial LM company pivoting into scientific AI — Jumper knowledge invaluable, product breakthroughs need time + org alignment.

Bottom line

Anthropic multi-year bet: AI compresses drug discovery timelines at bedside — not just benchmarks. Claude for Life Sciences infra + Coefficient Bio + Karpathy pre-training + Jumper = coherent strategy. AlphaFold-scale breakthrough unknown — pieces more carefully assembled than most realize.

Biotech teams running Claude Code + MCP on long genomics/regulatory automation workflows: heavy Agent sessions on dedicated cloud Mac eliminate 16GB laptop swap thrashing, deliver stable SSH + Keychain isolation for macOS toolchain. NodeMini rental pricing + help center for setup. Corporate trajectory: Anthropic IPO guide.

info

Disclaimer: Data as of June 30, 2026 pre-livestream. Post-event update recommended for product announcements, Jumper official role, Fable 5 restore, new pharma partnerships. Not medical or investment advice.

FAQ

Life Sciences FAQ

The Briefing: AI for Science — livestream 30 июня 2026 (10:00 AM PST) из San Francisco. Anthropic showcase Claude for Life Sciences + demos/case studies от Novartis, BMS, Genentech и др.

John Jumper co-led AlphaFold 2 в Google DeepMind, Nobel Prize Chemistry 2024 в 39 лет. 19 июня 2026 объявил move в Anthropic после ~9 лет в DeepMind. Specific role undisclosed.

Internal testing: Claude Mythos 5 ~10× acceleration key drug design steps. Fully autonomous: viable candidates для 9/14 protein targets zero human assistance.

На 30 июня 2026 Mythos 5 только у ~100 vetted US orgs после partial restore 26 июня. Fable 5 offline с 12 июня. Stable Claude Code + MCP dev env: цены аренды cloud Mac.

Launch Oct 2025 — enterprise stack для pharma/biotech R&D. Connectors: Benchling, PubMed, 10x Genomics, Medidata, ClinicalTrials.gov — literature synthesis, hypothesis gen, clinical trial drafting, regulatory submissions.

Apr 2026 ~$400M all-stock. Sub-10 team из Genentech Prescient Design — protein design + biomolecule modeling, critical bridge LLM assistant → AI drug discovery engine.

NovoScribe на Amazon Bedrock + Claude + RAG templates. CSR writing time −90%. Expanding → device protocols, patient materials, full CTD automation.