Anthropic «AI for Science»
John Jumper · drug design ×10 · pharma CEO on stage

Mythos 5 benchmarks · Novo Nordisk −90% CSR · Coefficient Bio $400M · export control tension

Anthropic AI for Science event 2026 life sciences briefing
TL;DR: 30 июня 2026 Anthropic проводит крупнейший life sciences moment — The Briefing: AI for Science в San Francisco (10:00 PST). Нобелевский лауреат John Jumper (AlphaFold) только что ушёл из DeepMind в Anthropic. Claude Mythos 5 в internal tests ускоряет drug candidate design на ×10. Novo Nordisk сократил время написания Clinical Study Reports на 90%. В этом разборе: ① speakers включая CEO Novartis, BMS, Genentech; ② arc Jumper и strategic timing; ③ 18-месячный build-up и Claude for Life Sciences connectors; ④ Mythos 5 benchmarks и pharma deployments; ⑤ export controls, Coefficient Bio и шестишаговый runbook.
01

Что такое event «AI for Science» от Anthropic?

The Briefing: AI for Science — live stream из San Francisco 30 июня 2026 в 10:00 PST. Agenda: vision Anthropic для AI в scientific research, Claude for Life Sciences demos и customer spotlights от top pharmaceutical и biotech companies.

SpeakerRole
Vas NarasimhanCEO Novartis; board member Anthropic
Chris Boerner, PhDCEO Bristol Myers Squibb
Aviv RegevEVP & Head of R&D, Genentech
Lotte Bjerre KnudsenProfessor, DMSc; ex Chief Scientific Adviser Novo Nordisk
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 выходят на сцену AI-компании — это не marketing stunt, а signal о direction pharmaceutical R&D. Traditional drug development: 12–15 лет, cost >$2.6B; только ~10% clinical candidates доходят до approval.

01

Target ID lag: Mining миллионов papers раньше занимал months; AI сжимает до hours.

02

Compound design bottleneck: Wet-lab screening медленный и failure-prone; simulation ускоряет на orders of magnitude.

03

Regulatory documentation drag: CSR пишутся months — core pain point Novo Nordisk.

04

Compliance & explainability: Pharma требует auditable AI outputs; Constitutional AI path Anthropic fit лучше.

05

Cross-border access risk: Mythos 5 export controls создают real uncertainty для international pharma teams.

02

Кто такой John Jumper и почему его move в Anthropic — big deal

John Michael Jumper (b. 1985, Little Rock, Arkansas): Vanderbilt BS math/physics → Cambridge MPhil physics (Marshall Scholar) → University of Chicago PhD theoretical chemistry (2017). Через шесть месяцев после PhD — Google DeepMind, secret protein-folding project AlphaFold.

CASP14 2020: AlphaFold 2 достиг experimental-grade accuracy за hours, не months. Impact: 214M+ predicted structures; 2M+ researchers в 190+ countries. 2024 — Nobel Prize in Chemistry Jumper и Demis Hassabis (с David Baker, UW). В 39 лет — youngest chemistry laureate за 70+ лет.

19 июня 2026 Jumper posted: «After nearly nine years, I have decided to leave Google DeepMind and join Anthropic.» Hassabis ответил публично. Announcement за 11 дней до сегодняшнего briefing — choreographed или coincidence, но frames весь science narrative Anthropic. Role undisclosed; likely steering foundational biological AI, возможно «ClaudeFold»-class tool.

DateEvent
Oct 2025Claude for Life Sciences launch
Feb 2026Allen Institute & HHMI Janelia partnerships
Apr 2026Coefficient Bio acquired (~$400M)
May 2026Andrej Karpathy joins pre-training team
Jun 9, 2026Claude Fable 5 + Mythos 5 launch
Jun 12, 2026US orders Fable 5 & Mythos 5 offline
Jun 19, 2026Jumper → Anthropic announced
Jun 24, 2026Bloomberg: AlphaFold co-authors Adler & Pritzel may follow (unconfirmed)
Jun 26, 2026Mythos 5 partially restored for ~100 US orgs
Jun 30, 2026AI for Science briefing
03

Claude for Life Sciences: stack и реальные capabilities

Launch October 2025 — Claude Enterprise с MCP connectors в tools, которые scientists уже используют daily.

PlatformCapability
BenchlingDraft protocols, SOPs, informed consent из ELN/LIMS
10x GenomicsSingle-cell & spatial transcriptomics
PubMedSearch, retrieve, synthesize papers с citations
bioRxiv / medRxivPreprint access до formal publication
Open TargetsTarget identification & prioritization
MedidataTrial enrollment & site performance monitoring
ClinicalTrials.govCompetitive landscape из registries
Wiley Scholar GatewayFull-text journal access
BioRenderScientific figure workflows

Coverage: early discovery (literature, hypotheses, target ID) → preclinical (genomics, scRNA-seq QC, tox prediction) → clinical (FDA/NIH-aware protocols, enrollment monitoring) → regulatory (gap analysis, agency query responses).

Coefficient Bio (Apr 2026, ~$400M all-stock): Founders Samuel Stanton & Nathan C. Frey из Genentech Prescient Design. Goal: «ASI for Science.» Investor Dimension reported 38,513% IRR. Team merged в Anthropic healthcare под Eric Kauderer-Abrams — protein design и biomolecule modeling, bridge от assistants к настоящим drug-discovery engines.

04

Шестишаговый runbook после AI for Science briefing

01

Track live announcements: Jumper role reveal, Mythos 5 biology access expansion, Fable 5 restoration, new partners. Official: Claude for Life Sciences, Fable 5 & Mythos 5 report.

02

Map Mythos 5 access: As of June 30 — только ~100 vetted US orgs. International teams нужны compliance fallbacks — см. Fable 5 ban & alternatives guide.

03

Prioritize MCP connectors: Align Benchling, PubMed, Medidata integrations к вашему pipeline stage. Study NovoScribe RAG + expert-approved templates.

04

Dual-track model routing: Claude для heavy science tasks; OpenAI/Google/domestic alternatives для policy shocks. См. AI coding assistant comparison.

05

Watch international paths: Non-US researchers locked out из most capable science models Anthropic — plan regional AI stacks accordingly.

06

Lock stable Apple Silicon hosts: Bioinformatics pipelines и Claude Code agents требуют 7×24 uptime. Evaluate MESHLAUNCH cloud Mac rental — laptops sleep; Linux VPS lacks Metal и Xcode.

05

Насколько хорош Claude Mythos 5 в real science?

BenchmarkResult
Drug design speed~×10 на key steps, zero human help
Hit rate9 / 14 protein targets → strong candidates
WorkflowBinding sites → tool selection → design runs → self-recovery from failures
Target classesImmune checkpoints, growth-factor signaling, neurodegeneration, muscle disease, structural targets
AAV capsid (Dyno data)Beat dedicated protein language models
Hypothesis blind test~80% human preference vs prior Opus; E. coli antimicrobial target validated in lab
Autonomous genomics (1 week)138 species, millions of cells; custom ML model 100× smaller than recent Science model, better performance

Novo Nordisk (Ozempic maker): CSRs bottlenecked approvals. Built NovoScribe on Claude via Amazon Bedrock + RAG с domain-expert templates.

«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

Expanding to device protocols, patient materials, full CTD automation. Case study: claude.com/customers/novo-nordisk.

Other deployments: Sanofi, AbbVie, AstraZeneca, Genmab, Bristol Myers Squibb; plus Komodo Health (analytics) и Axiom (Claude Code + MCP tox prediction).

DimensionAnthropic edgePeers
Safety cultureConstitutional AI, regulatory-friendlyMixed approaches
Vertical stackConnectors + Coefficient Bio + JumperDeepMind has AlphaFold, different GTM
Enterprise lock-inNovartis, BMS, Genentech, Novo Nordisk depthExpanding competition

Export control cloud: Since June 12 — Fable 5 и Mythos 5 pulled для foreign-national access. June 26: Mythos 5 partially restored для ~100 US critical-infrastructure orgs. Fable 5 still down; negotiations ongoing. International pharma employees at showcased companies face operational uncertainty.

Can Jumper replicate AlphaFold at Anthropic? Genuinely unknown. AlphaFold имел DeepMind infrastructure, deep biology partnerships, crisp CASP benchmark. Anthropic — commercial LLM company pivoting в scientific AI; knowledge Jumper invaluable, но product breakthroughs need org-wide execution.

A

×10 design acceleration: Internal Mythos 5 benchmark, 9/14 autonomous hits.

B

90% CSR reduction: NovoScribe measured outcome, not marketing copy.

C

214M structures legacy: AlphaFold scale Jumper brings в product arc Anthropic.

Note: Written around June 30 livestream. Post-event updates recommended: product announcements, Jumper official role, Fable 5 restoration, new pharma deals. Sources: Anthropic, STAT News, Nobel Foundation, Novo Nordisk case study.

Life-science AI pipelines need stable compute — sleeping laptops, shared instances, Metal-less Linux VPS eat efficiency, которую обещает Mythos 5. Для 7×24 bioinformatics agents, parallel analysis, macOS toolchain work — MESHLAUNCH cloud Mac Mini rental обычно лучший production choice. См. Anthropic IPO guide для capital context (центр помощи).

FAQ

Live briefing 30 июня 2026 (10:00 PST) с Claude demos и pharma case studies. Cloud Mac pricing на странице тарифов.

AlphaFold 2 lead, Nobel Chemistry 2024 (youngest за 70+ years). DeepMind → Anthropic 19 июня 2026. Lecture: nobelprize.org.

~×10 на key steps в internal testing; 9/14 protein targets yielded strong candidates с full autonomous workflow без human assistance.

Только ~100 vetted US organizations на 30 июня. Fable 5 suspended с 12 июня. См. alternatives guide и центр помощи.

Anthropic, April 2026, ~$400M. <10 employees, mostly ex-Genentech computational biology, building AI drug discovery — brought protein design expertise in-house.