Most of this corpus assesses whether something will work. A few subjects raise a different question, which is whether the work should be done at all — and on those, informed people who agree about every measurement still reach opposite conclusions. Releasing a self-propagating engineered agent into a wild population. Driving an allele to fixation across a species. Putting a human neurodevelopmental gene into an animal that cannot be asked.
A brief on a subject like that can fail in two directions. It can adjudicate — reach a verdict the evidence does not support and dress it as a finding. Or it can flinch — cover the science and leave the reader unaware that a real argument is going on, which is its own kind of dishonesty. Adopted 15 September 2026, this page is the standard written to avoid both.
The order this happened in, stated plainly
The Institute’s first editorial position was adopted after the brief it governs was already published, and that page says so. This one was written before its briefs were, and the commitments below were given to their authors as constraints rather than applied afterwards as a review. The audit at the foot of this page is therefore a check that the constraints held, not a retrofit — which is the only version of this exercise worth anything.
The five commitments
1 · The subject is the evidence and the disagreement, not a verdict
Where a position is contested, the brief presents the strongest form of each side and states what evidence would discriminate between them. It does not conclude which side is right. This is the same assess-only commitment the military-AI standard makes, and it is load-bearing for the same reason: the four-flag system is worth something only while a reader can trust that a flag reflects evidence rather than the Institute’s preference.
2 · Factual disagreement and values disagreement are separated explicitly
This is the commitment that does the most work, and it is the one a reader should hold the Institute to hardest. Some disputes on these subjects are about facts and could be settled by a measurement: whether a caged result transfers to the field, whether a confinement design actually confines, whether removing a species from a food web does what the models say. Others cannot be settled by any measurement at all: who may consent on behalf of a population that crosses a border, whether irreversibility is a distinct category of harm, whether an animal’s altered experience matters morally.
Both kinds are real. Conflating them is how these arguments go wrong — each side presenting its values position as though a study would vindicate it, or dismissing the other side’s factual claim as mere values. So the briefs name which kind each dispute is, and mark the ones that no measurement resolves. Where a dispute is genuinely mixed, they say that too.
3 · No operational content
No construct designs, no engineering methods, no protocols, no parameters — nothing that would function as laboratory guidance. The line these briefs are written to: readable by a legislator, a regulator or a review board, and useless to someone building the thing. Where a section could not be written honestly without crossing that line, the brief is written around the governance question instead and says in its own prose that the technical specifics are deliberately absent, and why. That is a better brief, not a compromised one.
4 · Interested parties are flagged on every side
Programme funders, researchers with a line of work to defend, national regulators, and campaigning organisations all have structural reasons to overstate or understate — and on these subjects the advocacy is unusually well resourced in both directions. Figures are labelled with whose they are.
5 · Irreversibility and consent are stated plainly, not aestheticised and not amplified
A release that cannot be recalled, an allele that crosses a border without anyone’s permission, an animal that cannot be asked: these are facts about the subject and they are reported soberly, in the register the evidence supports. Neither the technical vocabulary that makes them abstract nor the rhetorical amplification that makes them lurid.
Why present rather than judge, and the argument against it
The case against this standard is serious and deserves its strongest form: on subjects where the stakes include irreversible ecological change or the moral status of a creature that cannot speak, an institution that lays out the arguments and declines to say anything is not being neutral. It is declining a responsibility it is unusually well placed to discharge, and its neutrality functions in practice as consent to whoever is already acting.
The Institute takes the other view, for a reason specific to what this corpus is. Its value to a reader is that the flags answer to evidence. On a subject where the Institute had a position, every flag would become a thing to be argued with rather than checked — and a reader who suspected the assessment of carrying an argument would be right to discount all of it, including the parts that are simply measurements. The judgement is that a reader is better served by a map of the disagreement they can verify than by a verdict they cannot.
This is a position about this corpus, not a claim that assessment is superior to advocacy. Organisations that take positions on these subjects do necessary work and several are cited in the briefs. The Institute is not one of them, and these pages say which it is.
What this standard does not license
It is not a reason to cover a subject at less than full depth, and it is not a shield. A brief bound by this standard still states what the evidence shows and what it fails to show, still says when a claim exceeds its source, and still reports an absent evidence base as a finding rather than an open question to speculate across. Declining to adjudicate a values dispute is not the same as declining to assess — and on all three of the subjects below, the assessment is unflattering to somebody.
What it binds, and what the audit found
Three briefs carry this standard: Self-Spreading Vaccines, Ecosystem Gene Drives, and Animal Cognitive Uplift. Each was checked against all five commitments before this page was adopted.
Commitment 3 holds in the strictest case. The self-spreading vaccines brief was written as a governance document throughout: it describes the single field release in the published record as a documented policy event and carries no vector rationale, no modification methods and no construct parameters. Every occurrence of the word “protocol” in it refers to the Cartagena Protocol, the Biological Weapons Convention’s absent verification protocol, or a pre-registered study design — none to a laboratory procedure. The gene-drives brief describes homing, resistance alleles and confinement architectures only as the vocabulary a regulator needs to read a risk assessment, and says so as a stated rule.
Commitment 2 produced the most useful section in the set. The gene-drives brief separates four factual disputes, each with a named discriminating measurement, from three that no measurement resolves — and flags the moratorium question as the mixed case that each side presents as purely its own kind.
Commitment 1 was tested hardest by the uplift brief, whose decisive experiment may be one that should not be performed. Rather than resolve that, the brief records the two independent arguments against running it, notes that the marmoset group’s own advance refusal to carry an experiment to term is itself a policy result, and leaves the question standing.
And the assessments are not soft. Each brief’s central finding is a hard one: that the body a research programme defers a release decision to does not exist; that the canonical “no resistance emerged” summary of the field’s headline experiment is wrong, and that of 117 modelling papers only twelve examine confinement at all; and that the decisive primate experiment has already been run once, is unreplicated seven years later, and is reported incompatibly by its own authors in two publications.
Scope
This standard binds the three briefs named above and any future brief whose subject combines a contested values question with an irreversible or non-consensual intervention. It sits on top of the practices that apply to the whole corpus — the four-flag system, the review policy, and the citation record — and replaces none of them. Briefs bound by it carry a link to this page in their header. A reader who thinks a brief has slipped into advocacy, in either direction, has a correction to report rather than a difference of taste, and the way to make it is to the Institute directly.