Genesis
The thread opened on 2 May 2026, prompted by Neil Pollock's announcement of After Hype: The Business of Taming the Digital Economy (Cambridge University Press, open access). The "shock absorbers not fire alarms" formulation triggered a connection to Cabantous and Gond's three mechanisms and to a field-level performativity register that had been latent in the PSF programme since the AMR draft.
The same thread surfaced a complementary architecture: production-side (Pollock and Williams) plus field-level (this paper) plus consumption-side (PSF). The Cambridge School of Government as institutional home was raised in the same conversation. The paper sits squarely inside the PSF research programme as a field-level extension, not a separate framework. Performativity is the connective tissue, not a competing theory.
Theoretical Architecture
The paper operates on three layers, each anchored in existing literature, each contributing a different piece of the regulatory-proxy story.
Cabantous and Gond's three mechanisms
Conventionalising stabilises a contested concept (AI risk, AI capability, AI safety) into a tractable measurement object. Engineering builds the instruments (benchmarks, evaluations, conformity tests, model cards) that render the object actionable. Commodifying scales the apparatus across firms, jurisdictions, and standards bodies until alternative ways of evaluating become institutionally inaccessible. The three mechanisms run concurrently in AI regulation, not sequentially.
Pollock and Williams 2026
"After Hype" documents how analyst firms and AR professionals built a private-sector proxy apparatus over three decades. The "shock absorbers not fire alarms" formulation is structurally the same observation the paper develops: a device that smooths the very signal it claims to flag. Transposed to regulatory science, the question becomes whether NIST, AISI, and EU AI Act mechanisms perform a comparable smoothing function on AI risk discourse, and whether that smoothing is visible to the regulators who reach for the instruments.
Felicity conditions as analytic spine
Following MacKenzie and Callon, a Barnesian performative effect requires felicity conditions: institutional recognition, instrument legitimacy, the absence of competing evaluative apparatus. The paper specifies the felicity conditions under which a regulatory proxy successfully produces the world it claims to describe, and the conditions under which it fails. PSF supplies the consumption-side companion: when the felicity conditions hold and the proxy succeeds, the regulator's evaluative capacity erodes in tandem.
Empirical Sites (Candidate)
NIST AI Risk Management Framework
Conventionalising at scale. The framework defines what counts as an AI risk and what counts as managing it. Currently the most influential single reference document for US AI governance practice.
UK AISI and US AISI evaluations
Engineering work. Capability evaluations and red-team protocols as instruments that render AI safety tractable. Both institutes are publicly committed to evaluation as the substrate of policy, which makes the engineering layer unusually inspectable.
EU AI Act conformity assessments
Commodifying the apparatus across 27 member states with binding force. The Act enters full force in August 2026 and the conformity-assessment infrastructure is being built around the same proxy logic the paper interrogates.
Model cards and capability evaluations
Industry-produced instruments now circulating in regulatory and procurement settings, with no clear felicity condition for when they constitute evaluation rather than marketing.
Comparative case: SEC and financial model risk management (SR 11-7, OCC 2011-12)
Mature regulatory proxy apparatus in adjacent domain. Useful as a benchmark for what regulatory proxy infrastructure looks like after three decades of institutionalisation. Provides historical depth that the AI cases cannot yet supply.
PSF Connection
Three layers of the same mechanism, asked at different sites:
Pollock and Williams
How the apparatus gets built. Documented for analyst firms in the digital economy. Sets the template the paper transposes.
This paper
How the apparatus performs the world it claims to describe, under what felicity conditions, with what consequences for state evaluative capacity.
PSF (existing)
What the manager (or regulator, or evaluator) loses when reaching for the apparatus instead of evaluating directly.
Treating performativity as the field-level mechanism preserves PSF's analytic specificity (the consumption-side erosion) while extending the framework into a regulatory and state-capacity register where the stakes are different and the accountability surfaces are public.
Distinct Contribution
What the paper borrows
Cabantous and Gond on the three mechanisms. MacKenzie and Callon on Barnesian performativity and felicity conditions. Pollock and Williams 2026 as production-side empirical scaffolding. Thornton, Ocasio, and Lounsbury on institutional logics for the regulatory and state-logic register.
What the paper contributes
Empirical specification of felicity conditions for regulatory proxy performativity in AI governance. Three-layer architecture (production, field-level, consumption) connecting Pollock and Williams to PSF through performativity as the connective mechanism. Cross-jurisdictional comparison (US, UK, EU) testing whether the apparatus performs differently under different state-logic configurations. Diagnostic vocabulary for regulators and oversight bodies to detect when the apparatus is producing rather than describing the regulated object.
Possible Venues
Organization Studies
Cabantous and Gond's home journal. Direct fit for the three-mechanisms argument extended to public-sector evaluation infrastructure.
Regulation and Governance
Closer fit for the state-logic and felicity-conditions cross-jurisdictional comparison. Smaller theoretical lift, more direct policy traction.
Social Studies of Science
STS angle. How regulatory science constitutes the objects it claims to govern. Fit with performativity tradition and AISI evaluation studies as artefacts.
Organization Science
If the paper holds the field-level mechanism argument tightly. Higher bar than Org Studies but plausible for the cross-jurisdictional structure.
Sequencing Contingencies
The OSP Perspectives mechanism paper and the IfM First Year Conference take priority through May 2026. The empirical phase begins in summer 2026. This paper is held until both clear, with revisit scheduled after May 19.
Scope risk: three jurisdictions, three mechanisms, three layers, plus felicity conditions, plus PSF integration. The paper needs a tight empirical anchor (probably one regulatory artefact analysed deeply) to keep the theoretical machinery from running away.
Timing risk: the EU AI Act enters full force August 2026 and the regulatory landscape is moving. The paper benefits from being timely but suffers if written before the apparatus has been deployed at scale. Probable window: late 2026 or early 2027.
Co-authorship: a Cambridge School of Government collaboration is institutionally valuable but adds coordination overhead. Better to draft the pitch solo and approach potential co-authors with a defined frame than to start with an open conversation.
Watching List
Pollock and Williams 2026, After Hype: The Business of Taming the Digital Economy (CUP, open access). Read in full when scheduled. Cabantous and Gond, three-mechanisms papers, reread for direct application to public-sector evaluation. AISI and NIST publication streams for regulatory artefacts to analyse. EU AI Act implementation acts and conformity assessment guidance as they appear. Cambridge School of Government working papers and seminar series.
Inputs Needed Before Commitment
- Read Pollock and Williams 2026 in full (currently working from LinkedIn excerpt and Cambridge UP page).
- Map Cambridge School of Government research streams against the paper's frame, identify natural co-authors.
- Draft a 500-word pitch for Florian and Michael to test fit with thesis architecture.
- Decide whether the paper sits inside the thesis (as a field-level chapter complementing the consumption-side empirical work) or outside it (as a co-authored extension).
- Decide single empirical site (probably EU AI Act conformity assessments) versus comparative across three jurisdictions.