"Anthropic's Wet Lab, Artificial Wombs in 24 Months, and Scalable Gene Editing" — Moonshots

Why this is in the vault

This episode is the direct continuation of two running vault threads: Anthropic's documented pivot into wet-lab biology (first flagged via the bioweapon-risk disclosure in [[2026-09-17-moonshots-frontier-labs-slow-down-openai-ipo-anthropic-bioweapon]]) and the Colossal Biosciences de-extinction/artificial-womb trajectory tracked since [[2026-04-07-moonshots-ep-synbio-ben-lamm-colossal]]. Ben Lamm gives a specific, dated technical forecast (24-month extrauterine mammalian gestation; a 300→1,000 gene-edit scaling curve) and confirms Anthropic has moved from "intelligent middleware" to an actual CRISPR-adjacent wet-lab capability — a concrete AI-company-into-physical-biology data point worth tracking against the L5 north star and RDCO's "where does agent capability actually land next" thesis.

Episode summary

Recorded live at Moonshots Live 2026, Peter Diamandis (with co-hosts Dave Blundin and Alex Wissner-Gross) interviews Colossal Biosciences CEO Ben Lamm across de-extinction philosophy, Colossal's use of AI foundation models (OpenAI and Anthropic as early-access partners), scalable multiplex gene editing, and a specific 24-month forecast for fully extrauterine mammalian birth. The conversation closes with Lamm reacting to Anthropic's newly announced biology-lab/CRISPR-like capability as validating, not threatening, Colossal's bet on AI-accelerated synthetic biology.

Key arguments / segments

• [00:00-00:03] Cold open on the woolly mammoth project status and why de-extinction became Lamm's "billion-dollar project" despite zero synthetic-biology background going in — framed as a non-expert-advantage argument (ask the dumb questions of the smartest people in the room).
• [00:04-00:08] Colossal's AI stack: OpenAI and Anthropic as early-access partners used first as "intelligent middleware" (lab-ops/Jira integration, scientist workflow tooling) before models were capable enough for actual experiment design; Lamm says Colossal is deliberately "cautious" about investing in AI lab-automation.
• [00:07-00:09] References Anthropic's (then-recent) biology-lab announcement alongside Ginkgo Bioworks. Lamm frames near-term AI-in-biology value as literature unification and reproducibility (a 40-60% experiment-replication failure rate across published mouse studies, caused by inconsistent gene naming/classification) — not yet direct sequence-to-outcome prediction. Introduces Colossal's "Noah's Ark 2.0" global biorepository and the claim that "the dataset is more valuable than the model."
• [00:09-00:19] De-extinction philosophy and ethics: species-selection criteria (ecological role, indigenous-community input, conservation/educational value), the Stewart Brand "we are as gods" framing, and Lamm's explicit statement that Colossal does not apply its technology to humans.
• [00:19-00:22] The gene-editing scale curve: 20 edits was an industry-celebration milestone roughly a year ago; Colossal now consistently exceeds 300 edits at >90% efficiency and is testing into the thousands, roughly tripling edit count annually. Lamm expects DNA synthesis to eventually overtake multiplex editing as the dominant method.
• [00:22-00:23] The headline forecast: "within 24 months we at Colossal will have fully extrauterine mammalian birth — from conception to birth, without a surrogate mother." Explicitly framed as a technology-readiness claim, distinct from the separate societal/ethical/regulatory question of human application.
• [00:23-00:27] Next "iPhone moments" for Colossal: another de-extinction reveal, the artificial-gestation milestone, and an unannounced third project. Woolly mice case study: the public-facing phenotype was produced via 8 gene edits delivered in a single dose.
• [00:27-00:34] Audience Q&A: moonshot scope (vertical vs. horizontal problem-solving), ethical responsibility for ex-utero-born animals (social/emotional needs, herd reintroduction), elephant-conservation tech partnerships (Botswana's Elephant Havens, AI/satellite/drone migration tracking), and the mammoth-rewilding climate case (grass vs. tree carbon/albedo effects — contested even among Colossal's own scientists).
• [00:34-00:37] Announces AstroMech, a new Colossal-adjacent company building "inflectional models" — species/clade-level predictive layers meant to sit on top of (not replace) foundation models, likened to a hypothetical "biology Mythos." Lamm's own biggest-moonshot pick: reversing a projected 50% biodiversity loss over the next 25 years via government-funded global biorepositories.
• [00:37-00:39] Celularity/Lifebank USA placenta stem-cell banking; a Dubai biorepository partnership; a new U.S. Fish & Wildlife Service partnership under Secretary Zinke's conservation directive.
• [00:40-00:42] Closing exchange (prompted by a Cathie Wood question relayed backstage): Lamm's reaction to Anthropic's biology-lab/CRISPR-like announcement — calls it "a huge confirmation," evidence AI companies are converging on biology as "one of the most acceptable uses" of the technology. Declines to give a numeric disaster-probability estimate, says the media is "exaggerating" biosecurity risk while conceding "scary moments" are likely ahead.

Notable claims

• "Within 24 months we at Colossal will have fully extrauterine mammalian birth — from conception to birth, without a surrogate mother." [~00:22] — a dated, falsifiable forecast; worth a 2028 check-in.
• Gene-editing scale: 20 edits (industry benchmark, ~1 year ago) → 300+ edits at >90% efficiency (current, consistent) → testing into the 1,000-edit range, roughly tripling annually. [~00:20-21]
• Woolly mice phenotype achieved via 8 gene edits delivered in a single dose. [~00:26-27]
• 40-60% experiment-replication failure rate across published mouse studies, attributed to inconsistent gene naming/classification — framed as the near-term AI-in-biology opportunity (unifying terminology/literature search), not sequence-to-phenotype prediction. [~00:06]
• "I think the dataset is more valuable than the model itself." [~00:09]
• Colossal projects 50% global biodiversity loss over the next 25 years; argues this requires billions in government biorepository funding, not philanthropy/zoos alone. [~00:36]
• Colossal states it does not apply its technologies to humans — Lamm attributes this partly to FDA/regulatory friction and investor/governance differences, not purely ethics. [~00:11]
• Anthropic's biology-lab/CRISPR-like capability announcement (referenced, not technically detailed in this episode) is read by Lamm as validating — "a huge confirmation" that frontier AI labs see biology as a socially acceptable application of their technology. [~00:41]

Guests

• Ben Lamm — Founder/CEO, Colossal Biosciences (de-extinction, synthetic biology, artificial wombs); also founder of the newly announced AstroMech (biology-prediction "inflectional models") and connected to Celularity/Lifebank USA (placenta stem-cell banking).

Hosts: Peter Diamandis with co-hosts Dave Blundin and Alex Wissner-Gross ("the Mates").

Mapping against Ray Data Co

Medium-strong. Two load-bearing threads for RDCO: (1) the Anthropic-into-biology data point is a direct, concrete instance of a frontier AI lab moving past chat/agent interfaces into physical-world wet-lab capability — relevant to how RDCO reads "where does agent capability actually land next" for the L5 north star, independent of the biotech domain itself. (2) Lamm's "the dataset is more valuable than the model" framing, and the sequencing where foundation models were useful as workflow middleware well before they were useful for actual experiment design, is a reusable pattern for vertical-AI positioning arguments generally (including phData-adjacent domain-data-moat thinking) — the model only earned its keep in this vertical once it could handle a messy domain-specific reproducibility problem (inconsistent gene naming), not raw literature synthesis. Weaker on direct applicability: no architecture, pricing, or deployment detail here transfers to RDCO's own build work.

Related

• [[2026-09-17-moonshots-frontier-labs-slow-down-openai-ipo-anthropic-bioweapon]] — prior vault entry on Anthropic's bioweapon-risk disclosure; this episode is the "AI-into-biology, framed positively by a practitioner" counterpoint to that risk framing.
• [[2026-04-07-moonshots-ep-synbio-ben-lamm-colossal]] — first Ben Lamm/Colossal appearance in the vault; establishes the synthetic-biology/de-extinction baseline this episode updates.
• [[2026-04-19-moonshots-artificial-wombs-species-development]] — prior artificial-womb clip (northern white rhino case, "productionized species development" framing); this episode supplies the dated 24-month forecast that clip's thesis was building toward.
• [[2023-07-06-moonshots-ep53-ben-lamm-de-extinction]] — earliest vault entry on Lamm/Colossal, useful for tracking how far the gene-editing scale curve (20 → 300+ → testing 1,000 edits) has moved since.