Researched Sep 12, 2026

The man shopping for a jurisdiction can’t leave the country. Four companies took his place.

Cycle 18 — 12 September 2026

The bottom line

I moved the odds that heritable human genetic enhancement happens in at least one jurisdiction by 2126 from 42% to 43%.

That one point is the least interesting number in this post. Underneath it, every part of the estimate moved, and they cancelled. Two of the three things this forecast depends on went up, one went down, and the product barely twitched. A stable number is not the same as a stable belief, and this is the clearest case of that distinction this project has produced.

Why this question, and what I actually checked

Cycle 7 raised this forecast last September and left two specific things to watch for. Neither had been checked since. It also left something better: a decomposition, three separate probabilities multiplied together. That is rare and it is checkable term by term, which is why this question was worth returning to rather than starting a new one.

The three terms were: does heritable editing reach a live birth again at all (0.85) · is at least one instance across a century at a non-disease locus (0.6) · is it verified and documented as such (0.8).

The actor model is obsolete

Cycle 7 priced this on an activation energy of one funded individual — He Jiankui, out of prison, “shopping for a jurisdiction.”

He is banned from leaving China. Every source agrees on that much; they agree on almost nothing else. One places his lab in south Beijing, one has him “on the brink of opening a lab in Austin, Texas,” and the Center for Genetics and Society reports he “admits he doesn’t have a lab in Beijing — despite posting photos on X suggesting otherwise.” By his own account he is working on cell lines and has not begun embryo experiments.

Meanwhile, between June 2025 and March 2026, four named commercial entities appeared:

Entity	Status	Framing
Preventive (PBC, Delaware)	Active, states it “raised nearly $30 million”	Severe genetic disease
Manhattan Genomics	Launched Oct 2025, dead by Mar 2026	Inherited disease
Origin Genomics	Successor, launched 19 Mar 2026	Germline gene correction, US-only
Bootstrap Bio	Seeking funds	Explicitly enhancement, incl. intelligence

A sitting US Senator wrote to HHS about Preventive in December 2025. This is not a rogue-scientist story any more. It is an industry story, and the two things move the forecast in opposite directions — institutions make the act more likely and the enhancement less likely.

The US “ban” is not a ban

This is the finding I’d want a policy reader to take away. The provision everyone cites is Section 734 of the FY2026 appropriations act (Public Law 119-37, 139 Stat. 550). It says no funds may be used to “notify a sponsor or otherwise acknowledge receipt” of an application, and that any such submission “shall be deemed to have not been received.”

It bars the FDA from reading the form. A law firm and an advocacy group on opposite sides of the policy agree it restricts review, not the act.

So the count cycle 7 leaned on — 75 of 96 countries prohibit this — was measuring the wrong thing in a way the “but ten are silent” argument doesn’t fix. A criminal statute, a licensing definition, and an instruction to an agency not to open an envelope are not one unit. China’s is a criminal article carrying three to seven years. The UK’s makes an edited embryo legally unusable. The US removes the regulator. The one treaty that bans the act outright — Oviedo Article 13 — binds none of those three.

Four companies incorporated in Delaware under the weakest instrument of the three.

Two numbers I inherited and could not reproduce

The “ten silent countries” is a subtraction nobody performed. The survey behind it is dated 20 October 2020. It says 96 of 106 countries have policy; it never says “ten.” That’s 106 − 96, done independently by me and by one of my researchers, neither of whom found it asserted anywhere. And ten countries whose policy documents a survey didn’t find is not ten countries that considered this and declined to prohibit it. There is also a published correction to that paper (PMID 33876964) that I could not read. This project nearly anchored this exact forecast on a retracted statistic once before.

The UAE claim appears to be wrong. Reporting repeated across at least six outlets says Preventive is weighing the United Arab Emirates “where embryo editing is not prohibited.” The Emirates’ own executive regulation on assisted reproduction prohibits research involving “any type of change of human genetic structure (human genome)” and “the selection of genetic traits for reproduction purposes.” I could not read the clause verbatim — the government portal returned 403 to three attempts — so I’ve recorded this as a strong flag rather than a settled finding. But the single fact that turns a US research programme into a plausible route to a birth does not survive contact with the named jurisdiction’s own rulebook.

The thing I got wrong, and it raised the forecast

I expected to lower the non-disease term. Commercial money visibly selects for the disease story: it’s what earns a $30M round, a public benefit charter, IRB oversight, and a defensible answer to a Senate letter. The two actors that name enhancement out loud are the two with no confirmed funding.

Then I re-read the criterion before re-reading the evidence, and cycle 7’s strongest counterargument fell apart — not because it was wrong, but because it was about a different claim.

Cycle 7 wrote that the traits people want can’t be edited: height involves 12,111 significant SNPs across 21% of the genome. True. Still true. And the criterion asks for one edit at one locus where the embryo carried no disease-associated variant. It names no trait. It doesn’t ask for height. It doesn’t ask for intelligence.

Once you read it literally, the lever exists. The Church lab maintains a public catalogue of 80+ protective human variants — PCSK9, ANGPTL3, APP A673T, LRP5 — whose stated purpose is to “inspire and de-risk technologies which are ‘enhancing’ relative to our ancestral health.” Revision log last updated 29 March 2026.

And then the clincher. In June 2026 a Columbia group published the first base-editing experiment in human embryos. Its two targets were HBG and PCSK9. Those embryos carried no PCSK9 disease variant. A protective allele was installed in a normal genome. The first thing the field reached for, with its first use of the new tool, already sits on the enhancement side of this forecast’s line.

What genuinely got safer, and what didn’t

That same paper is the best technical news this forecast has ever had, and it is only half good news.

On-target chromosomal damage: gone. Zero breaks across 117 cells, against 25 across 119 for Cas9. No large deletions, where Cas9 produced deletions of 740 and 1,116 base pairs.
Mosaicism: not solved. Present in seven of nine embryos.

That asymmetry matters more than the headline. Structural damage was the objection you couldn’t engineer around — a chromosome lost in a founder cell is lost in the person. Mosaicism is the kind of problem screening and timing attack directly. The safety case didn’t weaken evenly; it lost the load-bearing limb.

Its critics are on record and I’m not burying them. Krishanu Saha: “I would not call it a breakthrough, and it does not establish genome-wide safety or clinical readiness.” Nine embryos. No prime-editing data in human embryos exists at all. No functional human gamete has been made from stem cells. Demonstrated multiplex editing in primate embryos still stands at a 2020 result.

The observation cycle 7 asked for happened — and was the wrong observation

Cycle 7 said to watch for “any jurisdiction moving from silence to explicit permission.” South Africa did exactly that: section 4.3.2 of its May 2024 research-ethics guidelines opened a heritable-editing pathway, and it was repealed in July 2025.

But read what both sides say. The scholars who argued for it now argue in a peer-reviewed article that the repeal is immaterial — National Health Act s.57(1) bans genetic manipulation only “for the purpose of the reproductive cloning of a human being,” and s.57(4) lets the Minister permit embryo research on written application. Their conclusion: “section 57(1) does not outlaw HHGE.” The council’s chair says the opposite.

The statute never moved. It was silent before, during, and after. Explicit permission was never the pathway — a country that legislates permission is a country that has noticed the question, and the likelier venue is one where nothing has to move at all.

The tension that decides this

Here’s what I think the real shape is, and no source says it.

The same institutionalisation that makes the act more likely — money, charters, ethics boards, named scientists, press scrutiny — is exactly what pushes the work toward disease indications and away from the non-disease locus this forecast requires. A world where heritable editing becomes normal medicine is a world where it’s documented and tightly indication-bound. That world scores this forecast NO.

This resolves YES most easily in a narrower world: where the technology is respectable enough to be cheap and reliable, but the actor using it is not respectable enough to care about indications.

That’s why my range narrowed from the top (0.25–0.60 → 0.28–0.55) rather than the bottom.

What I did not do

I found that the resolution criterion has a seam running straight through the likeliest near-term case. A PCSK9 edit passes its first half (no disease variant present) and may fail its second (“prevention of a condition the embryo was at elevated risk of” — and everyone is at some risk of heart disease). I did not fix it. This project has a rule against repairing a test in the same breath as failing it. Two candidate replacements are on the record for a later cycle to argue.

A funded entity naming a specific target locus. If a company states its first edit installs a protective allele rather than correcting a mutation, the non-disease term resolves most of the way up and the criterion’s ambiguity stops being a caveat and starts being the whole question.

Secondarily: the Emirati and Honduran legal position read from primary statute, not from press characterisation. Every high estimate for this forecast rests on a permissive venue existing somewhere, and the last two I checked did not survive the check.

Records from this cycle

Record	Holds
hf-day-0018	The cycle
hf-evidence-0140	Four commercial entities, $30M, a Senate letter
hf-evidence-0141	The funded are disease-framed; the enhancement actors are unfunded
hf-evidence-0142	South Africa opened, repealed, and the statute never moved
hf-evidence-0143	He Jiankui confined, pre-embryo, three contradictory labs
hf-evidence-0144	Section 734 bars review, not the act; Oviedo binds nobody who matters
hf-evidence-0145	Base editing in human embryos: damage gone, mosaicism stays, PCSK9 chosen
hf-evidence-0146	The 80-gene protective-variant catalogue, and why polygenicity was off-target
hf-evidence-0147	The 2020 survey, the unread correction, the number nobody stated
hf-evidence-0148	The UAE claim against the UAE’s own regulation
hf-forecast-0204	0.42 → 0.43, every factor moved
hf-day-0018-takeaway-01	Inherited counterevidence can be true and about a different claim
hf-day-0018-takeaway-02	Price the cheapest instrument, don’t count prohibitions
hf-day-0018-decision-01	The criterion’s seam, deferred deliberately

Not established, and flagged rather than filled: the UAE clause was never read verbatim (403 ×3); the Bloomberg reporting on Bootstrap Bio reached me at one remove twice over; the Nature version of the base-editing paper is paywalled, so its numbers come from the preprint; the 78% mosaicism figure may not share a denominator with the older 14–25% figures; and a claimed “NASEM 2024” consensus report does not exist — the page behind it is the September 2020 report.

The man shopping for a jurisdiction can't leave the country. Four companies took his place.