Rule 436(g) was a four-line exemption. It said that when a nationally recognized rating agency’s rating appeared in a securities registration statement, that rating did not count as an expert opinion under Sections 7 and 11 of the Securities Act — so the agency never had to consent to being named, and could not be sued as an expert for what its name was attached to. Section 939G of the Dodd-Frank Act repealed it on July 21, 2010. Moody’s had said six days earlier that, given the potential legal consequences, it could not consent to inclusion of its ratings in prospectuses and registration statements. Within about a day, Ford Motor Credit had pulled a planned asset-backed offering and written to the SEC asking for enforcement relief; the SEC granted it that same day and extended it indefinitely in November. Public asset-backed issuers went back to omitting the ratings — and the identity of the rating agency — from the documents that legally mattered.
Nothing about the names had changed. Moody’s name had been printed on those instruments for decades, in the marketing materials, in the trade press, in every investor’s mental model of what the paper was worth. The reputational stake was untouched; so was whatever credit the agencies were owed for their analytical work. What changed was that the name now cost something if the rating was wrong. The agencies’ answer, delivered in under forty-eight hours, was to withdraw the name.
That reaction is unintelligible on the two accounts of attribution we actually use. The first says credit is owed: a name marks a contribution, and marking it correctly is a matter of desert. The second says credit is an incentive: name people and they will produce more of what gets named. Both accounts locate the whole function of attribution in the naming. On both, the repeal of 436(g) should have been a non-event — the names were already there, the credit was already assigned, the incentive was already running. Neither predicts an industry treating the printed name as suddenly intolerable.
The prediction succeeds if you take the mechanism to be something else. Attribution’s operative function is liability attachment: it is the device that keeps a claimant inside the window where the artifact can still cost them something. Desert and incentive are the story we tell about it. When the two came apart in securities law — a century of naming with a statutory shield against consequence — nobody noticed anything was missing, because the shield was invisible and the name was doing the visible work. The moment Congress refused the shield, the parties holding it demonstrated, in a single business day, which of the two they had been holding all along.
A skeptic will answer that this is a story about risk rather than about names: a regulatory classification changed, open-ended liability appeared, and firms fled legal ambiguity. But Section 11 liability attached to the consent to be named. Nothing prevented the agencies from rating the paper, publishing the rating, or being universally understood to have rated it. What they refused was the signature.
And if the name were valuable as credit while the liability were merely expensive, a market should have appeared — agencies charging issuers for consent, pricing the new exposure into the fee, as accountants and underwriters price theirs. None did. The name turned out not to be separable from the exposure at any price, which is what it means for the two to be one instrument rather than two. The sub-forty-eight-hour response indicates the dependence was not discovered in 2010. It was structural, and had been the whole time.
That also disposes of the gentler objection: that attribution is simply multi-functional — epistemic, coordinative, status-conferring, and liability-bearing — so the essay has found one function among several and promoted it. If that were right, the agencies would have traded off. They would have kept the epistemic and status functions, priced the legal one, and stayed on the page. A function abandoned in its entirety the first time it costs anything was not one of several.
Why this is not an enforcement gap
The strongest objection is a reclassification, and it deserves stating at full strength: nothing above requires abandoning the desert account. Attribution is about credit; the problem is that we assign it sloppily. Refine the taxonomy, specify contributions at higher resolution, require each named party to affirm what they actually did, and the gap closes. On this reading the essay has found an implementation failure and dressed it as a broken model.
This is not a hypothetical. It is the reform medical publishing already ran, twice, for exactly the reason the desert account gives.
In August 2013 the International Committee of Medical Journal Editors added a fourth criterion for authorship: agreement to be accountable for all aspects of the work, and to ensure that questions about accuracy or integrity are investigated and resolved. The ICMJE’s own explanation of the change names the motivation precisely — individual authors had been responding to misconduct inquiries by denying responsibility for the part of the work in question. Naming had failed to attach answerability, and the response was to write the answerability into the naming criteria. Note what the fix consists of: an agreement. A person who satisfies criteria one through three signs criterion four. Nothing fires if the signature turns out to be false.
The reason that matters can be stated precisely, and should be, because “exposure” is otherwise a word doing too much work. A consequence constrains behavior when it is (i) enforceable by someone other than the claimant, (ii) non-trivial in expected cost, and (iii) near enough in time to be foreseeable when the claim is made. Criterion four fails all three. The ICMJE assigns enforcement to the author group itself, and where the author group cannot agree, to the institution where the work was performed — explicitly not to the journal. The expected cost is close to zero, because the trigger is a misconduct inquiry that almost never opens. And the interval between signing and any inquiry is unbounded. The objection that these regimes are merely weakly enforced concedes the point in the wrong vocabulary. They are not weakly enforced. There is no external enforcer.
The prevalence data show what that produced. A 2024 systematic review and meta-analysis in Scientific Reports, pooling nineteen surveys in the health sciences, found honorary authorship at 18% when respondents were simply pointed at the ICMJE criteria — and at 51% (95% CI 47–56, fifteen surveys, 5,111 respondents) when respondents were asked to declare their coauthors’ actual contributions and those declarations were then compared against the criteria. The gap between 18% and 51% is the distance between what the naming regime asks people to affirm and what the work records show. Higher-resolution contribution reporting did not close it; higher-resolution contribution reporting is how it was measured.
The same reform is now running a second time on a fresh problem, and failing faster. An analysis of 5,114 journals and roughly 5.2 million papers found that about 70% of journals had adopted generative-AI policies, overwhelmingly of the disclosure-required type, and that the presence of a policy had no measurable association with AI adoption rates or with disclosure rates. Every major publisher’s policy converges on the same three commitments: AI cannot be an author, humans remain accountable, use must be disclosed. The first two are restatements of criterion four. The third is a new attestation. None of them names a consequence.
Two independent reform cycles, in the same institution, both of the form specify the naming better, both producing no movement in the behavior they targeted. That is the pattern a model-level failure produces. An implementation gap does not survive being fixed twice.
Which raises the question the failure alone does not answer: why this reform, twice, and not another. The answer is visible in the drafting. The ICMJE’s recommendations do not merely omit an enforcer; they name one and it is not the journal. Determining that everyone listed meets the four criteria is assigned to the authors collectively, and where the authors cannot agree, the matter goes to the institution where the work was performed — the text is explicit that this is not the role of journal editors. The body writing the accountability rule wrote itself out of enforcing it, in the rule.
I take that to be the selection mechanism, and mark it as inference rather than as something the ICMJE has said about its own motives: an attestation is the only form of accountability an institution can adopt at no cost to its own throughput. Every alternative — adjudicating contested authorship, auditing contribution claims, running an inquiry — is paid for in editorial capacity, which is the scarce good. Naming reform is not what reformers reach for because they are naive about incentives. It is what survives selection by the party that would otherwise pay, and that party is usually the one holding the pen. The reform is cheap in exactly the dimension that makes it inert.
What happens when the consequence is real
The contrast case is Sarbanes-Oxley. Sections 302 and 906 require the CEO and CFO to personally certify each periodic report — 302 civilly, 906 as a criminal statute carrying up to ten years’ imprisonment for a knowing false certification and twenty for a willful one. Those officers were already named on the filings. What SOX added was the price.
The behavior moved. Cohen, Dey and Lys, in The Accounting Review (2008), document accrual-based earnings management rising steadily from 1987 until SOX and declining significantly afterward. Pincus, Wu and Hwang re-examined the result over a much longer post-SOX window in the Journal of Financial Reporting (2022) and found the decline holds across the extended period, though the immediate-window estimate is sensitive to design choices.
And here the counterargument lands, because the same papers carry it. Real earnings management — cutting discretionary spending, timing production and sales — moved the other way, rising after SOX as accruals management fell. Firms substituted. Attaching a price to the certified number changed which instrument executives used to manage it, not whether they managed it. This is the limit of the argument: liability attachment does not make claims true. It makes the claimant bear the cost of their being false, and a claimant who can pay in another currency will. A further caution belongs here too. SOX was a bundle — certification, internal-control audits, board independence, a new audit regulator, arriving together — and no design in that literature isolates the certification requirement from the rest. The contrast between SOX and the ICMJE cycles is a contrast between regime types, not a clean estimate of what a signature-with-teeth buys.
Narrowed to what survives: consequence-attaching reforms change behavior in ways naming-only reforms do not, and they change it imperfectly, along the paths the consequence does not reach.
Three ways out, all bought in advance
Once attribution is understood as liability attachment, the ways of holding credit while escaping consequence stop looking like unrelated vices and resolve into one move at three addresses.
Invisibility. The ghostwriter cannot be embarrassed by the book. Removing the name removes the address the reckoning would be delivered to, which is why erasing credit is extractive in a way that has nothing to do with recognition being deserved.
Affiliation. The name stays and the institution absorbs the loss. This is what the honorary-authorship numbers describe: the senior figure whose presence on the byline supplies legitimacy the work has not earned, and whose department, tenure and grant portfolio survive the paper’s failure intact. Criterion four asks that person to agree to be accountable. The 51% figure indicates how much that agreement is worth.
Duration. The name stays, the liability is nominal and fully intact, and the reckoning is scheduled to arrive after the reward has been collected. The SOX substitution is this route in its documented form: cutting research spending or pulling sales forward moves the cost of the reported number into periods beyond the certifying officer’s tenure, and does it without touching a single accrual the certification covers. Nothing is hidden and nothing is misnamed. The signature is honest on the day it is signed. The exposure is simply timed to miss.
All three are purchased before any outcome is known, which is what makes them cheap. And none of them is detectable by looking harder at who is named, because in two of the three the naming is already correct.
What follows, and for whom
The prescription is narrow: stop treating naming reform as accountability reform, and spend the institutional effort on the consequence side. But the obvious version of that fails, and it fails by the mechanism this essay just described. Make a named individual answerable for post-publication integrity inquiries and the designation drifts to whoever can absorb it — the principal investigator for whom it is survivable, or the graduate student who is not in a position to decline. A person-attached consequence is offloadable, and the three routes above are exactly the means. Corresponding authorship would become the next honorary authorship, and the reform would arrive at the disease.
So attach the consequence to the thing that cannot exit. Journals already designate a corresponding author, and already issue expressions of concern. Couple the two: an integrity inquiry delivered to the corresponding author starts a published clock. When the clock runs out, a durable, machine-readable notice goes on the paper — not on any person — recording that the inquiry went unanswered. The notice adjudicates nothing about the science. It records a fact about answerability, in the one location none of the beneficiaries can leave behind, because the citation, the CV line and the grant renewal all continue to point at it. The invisible party, the shielded party and the departed party can each evade a personal sanction. None of them can evade a mark on the artifact they are still collecting on.
This is not a large consequence and it is not meant to be. It is the smallest change that converts criterion four from an attestation into something with an observable failure state, using machinery that exists, at roughly the price of a database field. It also requires no one to accept the model above.
Whom it serves: readers, replicators, and anyone downstream of a claim who currently has no way to discover whether it was ever answerable.
Whom it fails, stated here rather than in a footnote, because it is a structural feature of the prescription and not a caveat on it: everyone who cannot afford exposure. A pseudonymous writer under a government that jails critics. A whistleblower. And, inside this very design, a junior first author whose CV line carries a notice because a senior author declined to answer — routing the consequence to the artifact reduces that harm against a personal sanction and does not remove it, because the junior author’s stake in the artifact is the larger share of what they have. An exposure criterion tells you that a well-bonded liar has more standing than an unbonded truth-teller, and I do not believe that. I am committing to the claim that exposure is the operative variable in how attribution regimes actually behave. I am not committing to the claim that exposure should be the criterion of legitimacy, and the difference matters most exactly where the people are least able to argue for it. What would reopen the first commitment: a naming-only reform that demonstrably moves the behavior it targets, in a design that separates it from any concurrent change in consequences.
That gap is real and it is not going to be closed by anything in this essay. It is also not an argument for the status quo, which does not protect the vulnerable — it protects whoever already owns an exit.
The question worth putting to any disclosure regime is not whether it requires the truth to be stated. It is what happens, to whom, when the statement turns out to be false. Where the answer is nothing, what has been built is not accountability. It is a signature.
Open Questions
Ω_E — The isolation problem. Empirically resolvable. No existing design separates the SOX certification requirement from the rest of the 2002 bundle, so the size of the effect attributable to personal certification alone is unknown. A staggered-adoption setting — a jurisdiction that imported officer certification without the audit and governance provisions — would resolve it.
Ω_C — What “exposure” ranges over. Conceptually underspecified. Criminal liability, civil liability, retraction, professional censure, and reputational loss are all consequences, and the essay treats them as one variable. They almost certainly differ in effect and in who can absorb them. Specifying the index — which consequence, borne by whom, at what career stage — would split this into several distinct claims, at least one of which is probably false.
Ω_P — Exposure versus access. Structurally irresolvable by analysis. Whether legitimacy should track answerability, given that answerability is purchasable and unequally affordable, is a distributional judgment, not a finding. It belongs to whoever sets the standard, and the essay’s contribution is to make visible that they are setting it.
Forecast Register
# FORECAST REGISTER v1essay: the-byline-is-not-a-bonddate_written: 2026-07-25forecasts: - id: F1 hypothesis: "Attribution reforms that add or refine naming without attaching a consequence do not change the behavior they target; reforms that attach a consequence do." column: mechanism question: "In the most recent study published between 2026-07-25 and 2030-01-01 that estimates generative-AI text prevalence across scholarly journals grouped by AI-policy type, will estimated AI-assisted text prevalence in journals with disclosure-required policies be equal to or higher than in journals with no stated AI policy?" resolution_date: 2030-01-01 resolver: "Most recent peer-reviewed study or arXiv preprint reporting both figures for the same journal set. Resolves UNRESOLVED, not YES, if no study reports both. Phrased so that either direction is a positive finding and neither depends on a null being publishable." p_essay: 0.80 p_baseline: 0.45 reference_class: "Published evaluations of voluntary disclosure mandates in professional self-regulation, where roughly half report no reduction in the disclosed behavior." direction: fragility - id: F2 hypothesis: "Attribution reforms that add or refine naming without attaching a consequence do not change the behavior they target; reforms that attach a consequence do." column: magnitude question: "Will at least one of Elsevier, Springer Nature, Wiley, or Taylor & Francis publish, on its own author-facing policy pages, a named sanction (retraction, published correction, or submission ban) triggered by undisclosed generative-AI use in the absence of any other misconduct finding, by 2029-07-01?" resolution_date: 2029-07-01 resolver: "The four publishers' public author-guideline and research-integrity policy pages." p_essay: 0.35 p_baseline: 0.35 reference_class: "Publisher policy escalation from disclosure requirement to named sanction, 2015-2026; base rate low and this row carries no claimed skill over it — it is included to score the timeline estimate separately from the mechanism call in F1." direction: stability - id: F3 hypothesis: "Higher-resolution contribution reporting does not reduce honorary authorship, because it refines naming rather than attaching consequence." column: mechanism question: "Will a peer-reviewed study published between 2026-07-25 and 2030-01-01 report that journals mandating CRediT or equivalent structured contributor statements have lower honorary-authorship prevalence than journals not mandating them, at p<0.05?" resolution_date: 2030-01-01 resolver: "PubMed or Web of Science indexed peer-reviewed publication reporting a between-journal comparison of honorary-authorship prevalence by contributor-statement policy." p_essay: 0.15 p_baseline: 0.50 reference_class: "Published evaluations of structured-reporting mandates in research integrity, which report positive effects roughly half the time." direction: fragility - id: F4 hypothesis: "Higher-resolution contribution reporting does not reduce honorary authorship, because it refines naming rather than attaching consequence." column: magnitude question: "Will a systematic review or meta-analysis published between 2026-07-25 and 2030-01-01 report pooled contribution-based honorary-authorship prevalence in the health sciences at or above 40%?" resolution_date: 2030-01-01 resolver: "PubMed-indexed systematic review or meta-analysis reporting pooled prevalence under the contribution-declaration operationalization." p_essay: 0.75 p_baseline: 0.60 reference_class: "Prior pooled estimate 51% (95% CI 47-56), Scientific Reports 2024; baseline assumes mild regression toward reported declines in integrity-metric literature." direction: stability
cafebedouin.org 