By W.H.L., GPT-5.6 Sol, Claude Sonnet 5
Chapter 9 of On Gradual AGI
Publication Version v1.0.2 · September 16, 2026
Empirical chapter integrating Gradual AGI Series #10
Version note: Publication Version v1.0.2 synchronizes Chapter 9 with the published Chapter 8 v1.1.5 by naming the Reciprocal Theory of Participation and clarifying the prospective status of reciprocal Participatory Role Plasticity. The frozen Frame One denominator, coding, proposition verdicts, temporal analysis, non-veto findings, and Responsive Floor adjudication are unchanged.
Abstract
Chapter 8, now published as Participation: Subjects, Acts, and Consequences, defines Participation conceptually. This chapter asks which of those distinctions survive empirical contact. It integrates the 3M Taxonomy – Manifestness, Modality, and Materiality – with a frozen Act Form field and the Relational Grammar of Realization Orientation (RO), Foreclosure Position (FP), and Actor Dependence (AD). Only Act Form/Modality and RO/FP/AD were prospectively frozen; Manifestness and Materiality are secondary descriptive lenses. Accordingly, Proposition A tests the Grammar’s discrimination/nonredundancy, not the full 3M Taxonomy as an integrated instrument. The empirical core is Frame One, a bounded sponsor/project-announcement frame of 49 U.S. AI-related data-center projects, followed by a ten-case instrument-development pilot and a seeded 21-row confirmatory draw. At the August 28, 2026 cutoff, 14 confirmatory cases were mature and seven right-censored; the mature set contained 11 Present, two SearchedNotFound, and one Unknown.
The strongest result is structural rather than causal. Similar Act Forms preserve different RO, FP, and AD states in the executed coding, but that discrimination result is conditional on faithful application of the instrument because intersubjective reproducibility cannot be reconstructed from the surviving archive. Act, Consequence, and Realization remain empirically separable: cases stop at bounded Null, informational, procedural, and realization-relevant stages. Lancaster and Port Washington document non-veto association sequences, while Frontier provides the direct-authority comparator; at least one documented non-veto sequence is observed, but no frequency, effect-size, or generalizability claim follows and the aggregate Responsive Floor remains NotAdjudicated. A frozen temporal analysis does not identify a positive 2026 resistance-frequency effect. Out-of-frame developments through September 14 instead motivate a different prospective hypothesis: Participation may be becoming more anticipatory, institutionalized, and directed toward defining, conditioning, monitoring, and pacing realization. The formal extension rejects a universal Participation scalar or direct fifth argument in the Chapter 4 realization mapping and retains only typed partial transitions whose arrows can fail. Frame One tests only the human-side empirical apparatus; it does not estimate the full reciprocal human-AI configuration described by Chapter 8’s Reciprocal Theory of Participation, which remains conceptually admissible and prospectively testable rather than a result of this frozen frame.
Keywords: Gradual AGI; participation; Reciprocal Theory of Participation; Participatory Role Plasticity; 3M Taxonomy; Act Form; Relational Grammar; Conditional Non-Fungibility; Responsive Floor; non-veto pathway; empirical adjudication; realization
Scope note
The published Chapter 8 owns the conceptual architecture of Participation and situates the human-side Subject-Act-Consequence apparatus inside the broader Reciprocal Theory of Participation. Chapter 9 owns empirical adjudication: the 3M descriptive apparatus, frozen Act Form and RO/FP/AD coding, Frame One, consequence classes, pathway evidence, reliability, nulls, temporal findings, and aggregate Responsive Floor adjudication. Chapter 9 tests only the human-side operationalization; it does not recode Frame One to estimate reciprocal human-AI role configurations, and it does not reopen the published definitions merely because some tests are weak, unexercised, or unresolved.
Reader guide
The chapter escalates the evidentiary burden in one direction:
represent → discriminate → consequence → realization → classification → formal limit
Each step asks more of the evidence. A result that survives an earlier stage does not inherit the status of a later one. The principal reading rule is therefore the same as in Chapter 8: every forward transition may matter, but none is automatic.
9.1 From Conceptual Architecture to Empirical Contact
The published Chapter 8 ends with a deliberate handoff from conceptual admissibility to empirical survival. A conceptual architecture can distinguish Subject, Act, Consequence, Recognition, Uptake, Relational Grammar, Role Plasticity, and Responsive Floor while still failing under observation. Categories may prove redundant; coding may prove unreliable; proposed pathways may be absent; nulls may dominate; or the available evidence may be incapable of adjudicating a proposition at all.
Chapter 8’s reciprocal extension does not alter this empirical burden. It situates the human Participatory Subject within changing human-AI actor configurations and treats reciprocity as potentially instantiated through Participatory Role Plasticity rather than symmetrical subjecthood. Frame One was not designed to observe that reciprocal configuration directly, so Chapter 9 preserves the human-side operationalization and treats reciprocal role redistribution as a prospective research target.
Chapter 9 changes the burden from conceptual admissibility to empirical survival. It does not ask whether every idea in Chapter 8 can be illustrated. It asks how far the evidence can travel before the chain stops.
Participatory Act → Recognition → Uptake → Participatory Consequence → Possible Realization Change
The important word is possible. Empirical contact becomes informative precisely when one arrow fails.
9.1.1 A bounded empirical frame
The primary empirical domain is U.S. AI-related data-center development. Frame One contains 49 confirmed sponsor/project announcements. Inclusion was outcome-blind: an eligible human or human-organized subject had to perform an observable intentional act with an identifiable AI-related realization referent and sufficient evidence for the declared observation window. Project success, cancellation, approval, controversy, or absence of controversy could not determine inclusion. AI systems are not Frame One Participatory Subjects and were not coded with the human-side 3M/RO/FP/AD instrument. Frame One therefore does not measure the full reciprocal human-AI configuration and must not be retrofitted to do so.
The design deliberately favors a manageable denominator rather than an exhaustive national census. Ten cases were used for instrument development. A separate seeded draw produced 21 confirmatory rows. At the August 28 cutoff, 14 confirmatory cases were mature and seven right-censored. The mature set consisted of 11 Present cases, two SearchedNotFound cases, and one Unknown. Missingness states were preserved rather than converted into synthetic zeros.
9.1.2 Historical-coding preservation
Chapter 9 may reinterpret the frozen observations within the book architecture, but it may not rewrite them. Inclusion, Act Form, RO, FP, AD, preserved pathways, and evidentiary adjudications remain those of the underlying empirical study. Later terminology can clarify what a field means; it cannot retroactively change what was coded.
This is particularly important for the 3M Taxonomy. Modality has direct continuity with the prospectively frozen Packet-A field now called Act Form. Manifestness and Materiality are later descriptive applications to preserved acts. They therefore enter the book with unequal measurement status rather than being reconstructed as if the full 3M instrument had been prospectively validated.
9.1.3 Post-freeze rule
Developments after the Frame One research cutoff – and developments outside the data-center domain regardless of date – may appear only as explicitly labeled illustrations or prospective tests. They cannot alter N, the historical coding, proposition verdicts, reliability claims, headline rates, or Responsive Floor adjudication. Section 9.7 uses this rule to examine the still-unfolding 2026 environment without reopening Frame One.
The purpose of empirical contact is therefore not to rescue the framework by making every arrow succeed. It is to discover where the arrows stop.
9.2 Representing Participation: The 3M Taxonomy and Act Form
Before asking whether participation changes anything, the empirical design has to determine what kind of participation has occurred. This is a representation problem.
Chapter 9 operationalizes representation through two complementary structures:
3M Taxonomy = {Manifestness, Modality, Materiality}
Relational Grammar = (RO, FP, AD)
The two structures should not be collapsed. The 3M Taxonomy describes the act. The Relational Grammar locates the act in relation to realization.
9.2.1 Manifestness, Modality, Materiality
Manifestness asks how directly the act becomes observable in the relevant empirical record. Explicit petitions, public comments, legal filings, authorizations, published declarations, and formal organizational decisions are highly manifest; informal negotiation, behavioral adaptation, and internal deliberation can be harder to recover. Manifestness concerns observability, not importance.
Modality asks what kind of Participatory Act occurred. For reader-facing use in this chapter, Modality is represented by Act Form. The frozen source field was represented internally through the M1-M8 Participation Mode codes: M1 Expressive / Discursive; M2 Evaluative / Feedback; M3 Behavioral / Choice; M4 Deliberative / Consultative; M5 Contestatory / Claim-making; M6 Productive / Co-creative; M7 Adaptive / Reorganizational; M8 Authoritative / Constitutive. After this historical-method reference, the chapter uses Act Form as the reader-facing term. These are categories, not ranks.
Materiality asks whether the act bears substantively on the realization object under the predeclared relevance rule. Materiality is not assigned retrospectively from success. An act does not become material because it later changes realization, nor immaterial because it produces no detected consequence.
9.2.2 Act Form is not the Relational Grammar
Act Form is the reader-facing Modality field. It is not the RO/FP/AD tuple. The Relational Grammar consists separately of Realization Orientation, Foreclosure Position, and Actor Dependence.
Act Form ≠ (RO, FP, AD)
RO describes the direction of the act relative to the realization object. FP describes how open, constrained, foreclosed, or otherwise unsettled the relevant realization choice remains. AD describes how much substantive information or meaning would be lost if the actor were replaced by another otherwise comparable participant. Two acts can share Act Form while occupying different Grammar states; different Act Forms can sometimes occupy similar relational positions.
9.2.3 Orientation, foreclosure, and actor dependence
Act Form does not establish orientation. Contestatory acts can oppose, pause, redirect, condition, or report. Authoritative acts can support realization. F1-044 is the clearest supportive case: Hamby Water Supply Corporation performs an authoritative / constitutive act enabling project-specific water infrastructure. Thus Act Form does not imply RO.
FP is referent-specific rather than project-wide. A project can be physically advanced while an expansion, permit, infrastructure authorization, operating condition, mitigation obligation, or later governance rule remains open. F1-025 illustrates this with distinct acts coded FP-F, FP-C, and FP-O against different realization referents within one project.
AD is not authority. It does not encode formal standing, prestige, institutional power, correctness, or likelihood of success. F1-044 exercises genuine constitutive authority yet remains AD-M. Authority and substitution loss are different questions.
9.2.4 No participation score
The combined representation should not be collapsed into a universal scalar. Manifestness, Act Form, Materiality, RO, FP, and AD are heterogeneous objects with different types and no established common metric. Their relation to realization can also differ in sign and route. The empirical objective is therefore structured description before inference, not scalar compression.
9.2.5 Measurement status
The components enter Chapter 9 with unequal evidentiary status. Act Form has direct continuity with the frozen Packet-A Modality field. RO, FP, and AD were prospectively operationalized in outcome-blind coding. Manifestness and Materiality are later descriptive dimensions. The full 3M Taxonomy has not been prospectively validated as a unified instrument. The consequence for Proposition A is specific: its executed discriminant test rests primarily on RO/FP/AD against Act Form and ordinary descriptors; it is not a discriminant validation of Manifestness, Materiality, or the integrated 3M Taxonomy. Those two later dimensions remain secondary descriptive lenses pending prospective coding and reliability evidence. Section 9.3 therefore turns from representation to the empirical contract that supports – and limits – these distinctions.
9.3 The Empirical Contract: Frame One, Coding, and Reliability
The empirical contract fixes what can count as evidence before proposition-level interpretation begins. Four layers are kept separate: project membership, act coding, consequence coding, and realization adjudication. Later outcomes cannot back-propagate into earlier fields.
9.3.1 Denominator and confirmatory structure
Frame One contains 49 confirmed AI-related projects. A ten-case pilot developed the instrument; a separate seeded 21-row confirmatory draw was then frozen. At the cutoff, 14 confirmatory cases were mature and seven right-censored. Among the mature cases, 11 contained qualifying participation, two were SearchedNotFound, and one remained Unknown. Right-censored cases remain in the sample and remain right-censored.
The AI-relevance gate is likewise frozen. Twenty confirmatory projects had explicit project-side AI relevance; one additional case was retained on moderate-confidence contextual/public evidence; no project entered solely because a governance-side act used AI language. The gate therefore remains project-centered rather than outcome-centered.
9.3.2 Packet A before Packet B
Packet A captures the participation object before outcome information is opened. Its frozen reader-facing fields are Act Form, RO, FP, and AD. Packet B then codes consequence classes, realization status, linkage strength, and competing explanations. The separation is designed to prevent outcome knowledge from contaminating representation.
The consequence classes are non-exclusive. A single case may contain informational, procedural, institutional-return, distributional, realization-level, or other consequences at once. The framework therefore does not force cases into one terminal category.
9.3.3 Outcome blindness and anti-rescue
After outcome inspection, the study does not silently change the denominator, AI-relevance rule, random draw, observation window, missing-data state, Act Form definition, Grammar definitions, consequence vocabulary, realization outcome, contrast definition, precision gate, or censoring rule. A mechanism case cannot be promoted into a domain-level Responsive Floor merely because it is analytically attractive.
9.3.4 Reliability and archival loss
The registered reliability burden specified paired pre-adjudication coding, raw agreement, and Krippendorff alpha. One complete assignment-blinded X01-X08 return covering eight holdout cases survives. It is a single-coder raw record. It establishes that outcome-blind application was feasible, that unresolved FP values were permitted rather than forced, and that one coder could apply the instrument across heterogeneous acts. It does not establish inter-coder reliability or the consistency of the coding underlying the F1 discrimination cases. The matching pre-adjudication return cannot be safely paired with sufficient coder identity/class, packet/rule version, target, timing, complete row-level fields, X02 FP, and AD provenance. No defensible pairwise raw-agreement figure or registered alpha can therefore be computed from the surviving archive.
No remembered alpha, inferred agreement rate, or post-adjudication coefficient is substituted. The registered reliability statistic is unreconstructable under the present archive; future instrument validation remains possible only through new prospective independent replication, not retrospective repair of this study.
PropositionA_Reliability = NotReconstructable_ArchivalLoss
Relational Grammar: discriminant/nonredundancy evidence observed in the executed coding, conditional on faithful application of the instrument; intersubjective reproducibility cannot be reconstructed from the surviving archive, so instrument validation remains unresolved and requires prospective independent replication.
9.3.5 Measurement and evidentiary status
The apparatus therefore enters the results with intentionally unequal status.
Table 9.1. Measurement and Evidentiary Status of the Participation Apparatus
| Object | Present-study role | Status / boundary |
| Manifestness | 3M descriptive dimension | Secondary descriptive application; no prospective reliability or effect claim; not a Proposition A discriminant field |
| Act Form / Modality | Frozen Packet-A field | Primary act category; not a strength or outcome scale |
| Materiality | 3M descriptive dimension | Secondary descriptive application; assigned from predeclared relevance, not outcome; not a Proposition A discriminant field |
| RO / FP / AD | Frozen Packet-A Relational Grammar | Discrimination/nonredundancy observed, conditional on faithful application; reproducibility unreconstructable; prospective replication required |
| Consequence classes | Packet-B outcome representation | Non-exclusive; one case may occupy several consequence classes |
| Linkage grade | Packet-B evidentiary relation | Documentary linkage, not causal effect size |
| Responsive Floor | Domain-level realization classification | Documented association sequence observed; aggregate Frame One verdict NotAdjudicated |
This table is not a hierarchy of validity. It is a ledger of what each object is allowed to do in the present chapter.
9.4 Conditional Non-Fungibility Under Empirical Contact
Proposition A asks a deliberately prior question: do the Grammar fields preserve distinctions that ordinary descriptors would erase? The test is discrimination and nonredundancy, not prediction of consequence.
9.4.1 Same Act Form, different orientation
Thirteen of the 16 Present confirmatory primary signatures are M5 Contestatory / Claim-making. That concentration creates a demanding within-form test: if Act Form already carried the relevant realization structure, the M5 acts should be difficult to distinguish further.
They are not. F1-038 contains two public petition acts around the Meta El Paso project. Jessica Burnett’s May 3 petition sought a pause and reopening or redirection of the trajectory and is coded RO-C. Carlos Perez’s May 22 petition sought to stop the development and supplies RO-O. Calling both simply petitions against the project would erase the difference between conditional redirection and opposition.
same Act Form ⇏ same RO
9.4.2 Foreclosure is not project stage
F1-025 supplies the clearest within-project FP sequence. The primary July 2025 stop-project intervention addressed a substantially committed phase and was coded FP-F. A later zoning appeal targeted a constrained but reviewable route and was coded FP-C. A later phase-two comment entered an open Planning Commission review and was coded FP-O. One project therefore contains:
FP-F → FP-C → FP-O
The sequence does not imply temporal reopening of one identical decision. It demonstrates that FP belongs to the act-referent pair, not to the project as one global stage.
9.4.3 Actor Dependence does not encode authority
AD also preserves information that authority labels do not. Some contestatory acts depend strongly on the identity or situated knowledge of the actor; others can be substituted with less loss. Conversely, F1-044 is an authoritative institutional act but remains AD-M. Constitutive authority concerns what an institution is empowered to bind; AD concerns substitution loss. The two therefore need not covary, and AD cannot be read as a power or prestige scale.
AD ≠ authority
9.4.4 Sensitivity to the F1-044 reconstruction
F1-044 supplies the sole M8 / RO-S observation in the frozen 16-row primary-signature matrix. That matrix indexes one frozen primary Packet-A act per Present project; later within-project acts remain configuration-level evidence rather than substitutes for the primary signature. Excluding F1-044 therefore makes M8 empty in that primary-signature occupancy and narrows Act Form breadth. The study does not use later or secondary within-project acts to repair that frozen coverage. The leave-one-out sensitivity nevertheless does not remove the same-form/different-RO contrasts, the Lancaster FP sequence, or the AD distinctions. Proposition A’s nonredundancy result therefore does not depend on the reconstructed supportive row, although the M8 breadth claim does.
9.4.5 Verdict
Relational Grammar: discriminant/nonredundancy evidence observed in the executed coding, conditional on faithful application of the instrument; intersubjective reproducibility cannot be reconstructed from the surviving archive, so instrument validation remains unresolved and requires prospective independent replication.
This result is narrower than validation and stronger than illustration. The Grammar earns analytical cost at the discrimination level only conditional on the executed coding being a faithful application of the frozen instrument. The surviving archive cannot independently verify that condition across coders. The result is therefore internally coherent but not demonstrably intersubjective; prospective replication is required.
Discrimination also does not establish consequence. The next stage asks what happens after a qualifying act occurs.
9.5 From Act Form to Consequence: Where Conversion Stops
A theory of Participation becomes empirically useful only if it permits the path to stop. Chapter 8 therefore separated Act, Recognition, Uptake, Consequence, and Realization. Frame One exercises those distinctions directly.
Act → Recognition → Uptake → Participatory Consequence → Possible Realization Change
9.5.1 A bounded Null is a real result
F1-016 is the clearest bounded Null. A qualifying participatory act is observed, yet within the relevant primary window no attributable informational, procedural, institutional-return, distributional, or realization-level change is detected. The act is not reclassified out of Participation merely because no consequence follows.
Act ⇏ Consequence
9.5.2 Informational and procedural stopping points
Other cases travel farther but still stop before realization. F1-032 reaches an informational response without attributed realization conversion. F1-042 reaches formal procedure and adjudication without the requested realization change. F1-017 also reaches response without realization conversion. These cases make it impossible to treat recognition, receipt, or procedure entry as synonyms for substantive realization change.
Consequence ⇏ Realization Change
9.5.3 Uptake-positive and realization-positive sets
The mature Uptake-positive set is F1-009, F1-025, F1-044, and F1-048. The realization-positive set is narrower: F1-025, F1-044, and F1-048. F1-009 has strong informational and procedural consequences while realization attribution remains unresolved.
The narrowing matters because it shows why successful processing is not itself realization conversion. It also makes the three realization-positive cases analytically valuable: they reach realization by different authority structures and different directions.
9.5.4 Realization does not mean termination
Lancaster and Port Washington proceed under conditioning rather than termination. Frontier is an enabling case. A binary stopped/not-stopped outcome would erase all three.
The realization vocabulary therefore preserves conditioning, enabling, redirection, and other material changes even when the focal project continues.
9.5.5 Linkage strength
Across the 11 mature consequence-assessable Present cases, the strongest case-level linkage is L3 in six cases, L2 in three, L0 in two, and L4 in none. The grade scale is evidentiary: L0 = none/unassessable; L1 = temporal sequence only; L2 = procedural or substantive linkage; L3 = direct documentary attribution; L4 = direct attribution plus independent corroboration. Linkage grade is not a causal effect size. The absence of L4 means that no case combines direct attribution with independent corroboration; even L4 would not by itself be sufficient for mechanism identification without a design that materially reduces principal rival explanations.
9.5.6 Figure 9.1 – Where Empirical Conversion Stops
Figure 9.1 makes the chapter’s empirical logic visible. Observed stopping points include a bounded Null (F1-016), informational response without realization conversion (F1-032), formal procedure without requested realization conversion (F1-042), and realization-level conditioning or enabling (F1-025, F1-044, F1-048). A qualifying act can stop immediately; processing can occur without downstream conversion; consequence can stop before realization; and a small subset reaches realization-relevant change. No arrow is automatic, and linkage is not a transition probability or causal effect.
9.5.7 The result of stopping-point analysis
Frame One therefore directly exercises four nonimplications:
Act Form ⇏ fixed consequence
Act ⇏ Consequence
Consequence ⇏ Realization Change
Realization Change ⇏ termination
These stopping points prepare the harder question. If realization can be conditioned or enabled, must the participating actor itself possess binding authority?
9.6 Non-Veto Pathways and the Responsive Floor Test
The distinctive burden of Proposition C is not whether participation can matter when the participant already holds binding authority. Governance already has a language for that case. The harder question is whether participation can reach realization through a pathway in which the participating subject does not itself control the binding decision. Operationally, this chapter calls a pathway non-veto only when the participating subject lacks binding authority and a documented mediated sequence reaches a realization-relevant consequence through an actor or institution that does hold the relevant authority; lack of authority alone is insufficient. Section 9.6.4 states the definition formally.
Three mature cases have RealizationLevel = Present. One is a direct-authority comparator. Two are non-veto configurations. The small number prevents a domain-level effect estimate, but it is sufficient to test whether the non-veto pathway is empirically empty.
9.6.1 F1-044 Frontier: direct-authority comparator
Hamby Water Supply Corporation holds the relevant binding authority. Its project-specific authorization enables the Vantage water line, and Packet B records a realization-level consequence at L3. The act is nevertheless AD-M because Actor Dependence measures substitution loss, not power: replacing the institutional actor would lose some project-specific relation and authority context, but the authorization does not depend on one irreplaceable individual. Thus binding authority can coexist with moderate AD. This is direct-authority enabling, not non-veto evidence.
9.6.2 F1-025 Lancaster: mediated conditioning
Lancaster is weaker but structurally different. Residents and advocates do not possess binding authority over the focal project. Their participation enters municipal and developer processes followed by environmental, operational, reporting, community-engagement, local-benefit, and financial commitments, including a Community Benefits Agreement and later prospective governance. The project continues.
The strongest linkage is L2 at the configuration level. Developer strategy, municipal goals, zoning, bargaining, infrastructure constraints, fiscal incentives, and prior commitments remain visible co-determinants. The case is therefore mechanism-consistent mediated conditioning; the contribution of participation is not isolated.
9.6.3 F1-048 Port Washington: strongest documented non-veto sequence
Port Washington carries the distinctive burden most directly. Participating residents do not possess authority to approve annexation, create the TID, or bind Vantage to project terms. The City does. Yet the City later states that community input informed stringent data-center zoning and sustainability commitments, while a separate direct-legislation petition enters a future-TID ballot path. The focal realization proceeds under conditioning rather than termination.
Πresidents(-B) → CityB → G′ → Rz′conditioned
The strongest linkage is L3: the City directly attributes zoning and sustainability commitments to community input. Municipal authority, annexation/TID structure, developer strategy, fiscal bargaining, infrastructure, timing, and other actors remain plausible co-determinants. This is a documented L3 non-veto association sequence; the counterfactual effect remains unidentified.
Table 9.2. Realization-Relevant Cases and Authority Position
| Case | Authority position / observed path | Realization relation | Linkage | Defensible inference |
| F1-025 Lancaster | Non-veto: residents/advocates → municipal/developer process → CBA / prospective governance | Conditioning | L2 | Mechanism-consistent mediated pathway; contribution not isolated |
| F1-044 Frontier | Direct authority: water-line authorization → implementation; participant holds binding authority; AD-M denotes moderate substitution loss, not authority | Enabling | L3 | Direct-authority comparator; binding authority with AD-M; not non-veto evidence |
| F1-048 Port Washington | Non-veto: community input → City decision environment → zoning/sustainability conditions | Conditioning | L3 | Documented non-veto association sequence; counterfactual effect unidentified |
9.6.4 What the three cases establish
Two of the three mature realization-positive cases lack participant binding authority. For this chapter, a non-veto pathway requires both conditions: (1) the participating subject does not hold binding authority over the focal realization decision, and (2) a documented mediated sequence reaches a realization-relevant consequence through an actor or institution that does hold the relevant authority. Absence of binding authority alone is not enough; a powerless act with no downstream consequence is not a non-veto pathway. The observed cases reject only a narrow reduction: realization relevance cannot be equated with the participant directly exercising binding authority. With two non-veto realization-positive cases, the result is an existence claim about documented pathways, not a frequency, prevalence, effect-size, or generalizability claim.
9.6.5 Responsive Floor: two burdens
The Responsive Floor test has two different burdens. The association-sequence burden asks whether a traceable institutional-record sequence links a pre-specified Participatory Configuration to a material realization condition, including through non-veto actors. The aggregate/domain burden asks whether participation-sensitive realization can be adjudicated across sufficiently comparable cases.
The domain-level burden requires:
1. an outcome-independent frame and comparator;
2. determinate participation states;
3. pre-specified and adequate precision;
4. sufficient comparability, stratification, or modeling of major realization conditions;
5. evidence that Participatory Configuration adds explanatory value beyond those conditions; and
6. explicit separation of documentary linkage from counterfactual causal identification.
9.6.6 Split verdict
ResponsiveFloorTest_AssociationSequence = DocumentedAssociationSequenceObserved
ResponsiveFloorTest_Aggregate = NotAdjudicated
ResponsiveFloor_FrameOne = NotAdjudicated
The aggregate verdict is driven by the tiny silent comparator, heterogeneous realization conditions, configuration-level outcomes, and continuing right-censoring. NotAdjudicated does not mean NotSupported; the required inferential burden was not met.
9.6.7 F1-044 sensitivity
Excluding F1-044 removes the sole direct-authority realization-positive comparator and the sole primary M8 / RO-S signature. It does not remove the existence of mediated evidence: Port Washington remains L3 and Lancaster remains L2. Aggregate precision remains insufficient. F1-044 therefore strengthens comparison but does not create the non-veto result.
9.6.8 Out-of-domain illustration: Meta employee resistance
A separate 2026 episode at Meta illustrates why Participatory Consequence and final outcome must remain distinct. The Model Capability Initiative, introduced in spring 2026 to collect employee computer-interaction data for AI training, generated substantial internal resistance. By early June, that resistance had produced identifiable concessions: employees could pause collection for limited periods and some could request exemptions (Reuters, June 2, 2026).
The program was subsequently suspended on June 22 after Meta discovered that sensitive data collected through the initiative had been exposed internally. Meta explicitly attributed the suspension to the data-security incident rather than employee opposition (Engadget, June 22, 2026). Later reporting described an employee petition exceeding 1,800 signatures (Fast Company 2026), but the chronology should not be compressed into the claim that resistance caused the suspension.
employee participation → pause / exemption concessions
data-security failure → program suspension
Employees lacked final corporate authority, yet their organized participation was followed by documented modification of how the program operated. The later suspension had a separately stated proximate cause. The episode is outside Frame One because it concerns workplace AI development rather than the frozen data-center domain; it is illustrative only.
9.6.9 What survives the Responsive Floor test
Frame One contains at least one documented institutional-record association sequence linking non-veto participation to later realization conditioning, with Lancaster providing weaker supporting configuration evidence. This establishes existence within the frozen frame, not frequency or generalizability. The counterfactual contribution of participation remains unidentified, and the cross-case evidence is too sparse and heterogeneous to adjudicate the data-center domain as a Responsive Floor.
The frozen evidence therefore reaches realization but stops before domain classification. Section 9.7 asks a different temporal question: what has changed in the broader participatory environment during the still-unfolding year 2026?
9.7 Time, Resistance, and Ongoing 2026 Transformation
The temporal question in Frame One began with a comparatively simple hypothesis: was resistance to AI-related infrastructure becoming more frequent in 2026? The frozen data do not support answering yes. That result should remain intact.
But it does not exhaust the temporal question. By September 2026, participation around AI had changed visibly in its objects, institutional locations, and forms. Some developments fall outside Frame One’s domain; others occurred after its August 28 cutoff. They cannot be imported into the denominator or used to rescue the temporal result. They can motivate a different question: what if the more consequential 2026 change is not simply how often participation occurs, but where it enters realization and what it seeks to condition?
9.7.1 The frozen result: no identified increase in resistance frequency
Frame One records 23 focal broad-resistance episodes over 152.18 pre-2026 project-months, or 15.11 events per 100 project-months, compared with 10 episodes over 88.61 project-months in 2026 through August 28, or 11.29 per 100 project-months. The crude rate ratio is approximately 0.75; exploratory age-adjusted analysis likewise does not identify a positive calendar-2026 effect.
The temporal data are insufficient to identify a positive calendar-2026 resistance-frequency effect; uncertainty is too wide to establish either an increase or its absence.
Calendar2026ResistanceFrequencyIncrease = NotSupported
NotSupported does not mean resistance declined, remained constant, or lacked importance. The registered positive frequency hypothesis did not survive its frozen test. The test also measured only events per exposed project time, not institutionalization, actor composition, anticipatory intervention, authority structure, or participatory object.
9.7.2 Realization-specific concern intensified during 2026
Independent survey evidence sharpens the distinction. The Annenberg Public Policy Center reported that opposition to new local data centers rose from 49% in its February-March wave to 61% in a nationally representative survey of 1,320 U.S. adult citizens fielded June 16-July 19. Broader views of AI remained comparatively stable: 39% expected a negative national impact over the next decade, statistically unchanged, while 68% continued to say government was doing too little to regulate AI (Annenberg Public Policy Center 2026).
A September 14 update from the Gardner Food and Agricultural Policy Survey strengthens the inference. Comparable February and August 2026 waves found increased concern about electricity costs, water use, and conversion of agricultural land associated with AI/data-center development across urban, suburban, and rural respondents (Gardner Food and Agricultural Policy Survey 2026).
general attitude toward AI ≠ response to a concrete realization pathway
These surveys measure attitudes rather than observed participatory behavior. They therefore do not establish a Frame One frequency effect or show that concern converts into action. They suggest only that resistance may intensify selectively at concrete interfaces of realization; the attitude-to-action transition remains a separate empirical question.
9.7.3 Expert participation can reach institutional consequence
The September mathematics controversy supplies a different pattern. An open letter opposing Caltech’s AI-focused Mathathon drew hundreds of signatories; OpenAI withdrew its compute-credit sponsorship and organizers modified verification/publication requirements (The Next Web 2026; Tao 2026).
The sequence does not isolate the open letter as the sole cause. Its narrower relevance is that organized expert participation entered an institutional response and shifted the debate toward what should count as acceptable and valuable realization in mathematics.
organized expert participation → institutional response → modified realization conditions
9.7.4 From episodic objection to negotiated conditions
A September 9 agreement among the American Federation of Teachers, the United Federation of Teachers, and Microsoft established the National AI Safety & Privacy Standard for schools, covering training-data use, tracking, human oversight, transparency, data control, and contractual recourse; districts can incorporate the protections into Microsoft customer agreements (Microsoft/AFT/UFT 2026).
organized stakeholder participation → negotiation → binding realization conditions
The analytic point is its institutional form: organized stakeholder participation becomes part of the terms under which future deployment is permitted to occur.
9.7.5 Participation is moving upstream and inward
The 2026 record also complicates the image of participation as outsider action only. Employees, researchers, and executives can intervene from inside realizing institutions. This chapter calls that institutional location endogenous participation; location does not determine orientation or authority.
At the frontier-lab level, Dario Amodei argued for pacing capability advancement so safety measures can catch up, and Sam Altman subsequently supported pacing and independent evaluators (Amodei 2026; Associated Press 2026). These are not ordinary outsider objections: actors near the center of frontier realization are themselves contesting pace and safeguards.
9.7.6 Embedded external participation
Amodei’s proposal goes further: Anthropic committed to continuing, employee-like access for independent third-party safety evaluators, and Altman said OpenAI would adopt the same model (Amodei 2026; Business Insider 2026). The evaluator remains organizationally external but is deliberately placed inside the realization process.
endogenous authorization → embedded external participation → continuing oversight of realization
This is not a new 3M dimension or Grammar field. It describes institutional placement: an episodic act can become a standing capacity for repeated observation and challenge.
The arrangement also creates a second-order governance question about who guarantees evaluator independence. Institutionalizing Participation therefore does not eliminate Governance; it creates new design questions about authorization, information access, protection, recourse, and review.
9.7.7 The trajectory itself becomes contested
The apparent convergence around pacing and external evaluation should not be mistaken for consensus. On September 14, President Donald Trump rejected stronger AI guardrails and emphasized U.S. competitive advantage; Vice President JD Vance likewise expressed skepticism toward frontier companies asking government to regulate them (Associated Press 2026).
The relevant participatory object can therefore become the trajectory of frontier AI development itself: pace, safety thresholds, oversight, information access, and the balance between safety and competitiveness.
9.7.8 From acceptance or rejection to defining, conditioning, monitoring, and pacing
reactive → anticipatory
downstream → upstream
external only → external + endogenous + embedded external
accept / reject → define / condition / monitor / pace
These are not developmental stages. External protest does not disappear when negotiated standards or embedded evaluation develop. Nor does the chapter claim that participation has become uniformly more effective. The emerging change concerns configuration, not guaranteed consequence.
As frontier AI becomes more capable and its realization more consequential, Participation may become more institutionalized, more anticipatory, and increasingly directed toward the definition, conditions, monitoring, and pace of realization rather than toward simple acceptance or rejection.
Resistance may intensify selectively at concrete interfaces of realization – physical infrastructure, resource burdens, workplace systems, professional practices, and frontier-risk governance – even when generalized attitudes toward AI do not shift in parallel.
Prospective evidentiary status. The contemporary cases are sufficient to motivate these hypotheses, not to test them. Prospective support would require repeated, consistently coded observation windows showing that participation occurs earlier relative to decision closure and/or that a larger share of acts enters standing, negotiated, or embedded institutional routes or targets definition, conditions, monitoring, and pace. The trajectory claim should contract if those shares remain stable or fall under consistent coding, or if the apparent change disappears once domain composition and source visibility are held reasonably constant. The null is therefore not ‘nothing changes in 2026’; it is that no durable directional shift is observed under prospective measurement. Operational definitions and precision gates for that prospective test are deferred to a new protocol rather than inferred retrospectively from these illustrations.
9.7.9 Ongoing, unfinished
September 2026 is not a retrospective endpoint. The relevant institutions, capabilities, controversies, and participatory arrangements continue to change. The proper separation is:
frozen evidence ≠ ongoing observation
ongoing observation → prospective testable hypotheses
Frame One establishes what could be supported under a bounded design through August 28. The subsequent record shows why that result should not be mistaken for a completed account of Participation. The broader prospective possibility is that the division of participation between increasingly capable AI systems and humans may itself be changing: AI systems may enter more substantive realization functions while humans participate earlier, through more institutionalized positions, and in defining the conditions under which realization proceeds. Frame One tests only the human-side empirical apparatus, so this reciprocal temporal interpretation remains prospective rather than a Frame One result. That interpretation is consistent with the book-wide Chapter 3 term Ongoing, Unfinished Transformation. Chapter 11 later owns the dedicated treatment of feedback, recursion, and transformation over time. Whether the pattern generalizes remains an empirical question.
9.8 Formal Placement, Failure Conditions, and Unresolved Burdens
The empirical results permit a formal placement of Participation, but not a fitted model of its effects. Frame One establishes that participatory configurations can sometimes be linked to consequences, that consequences can sometimes reach realization-relevant conditions, and that both transitions can fail. It does not identify a universal transition probability, common-effect coefficient, or stable function mapping participation directly into realization.
The appropriate formal result is therefore a specification constrained by empirical contact, not an estimated model.
9.8.1 Participation as a typed partial transition
Let Πr[t0,t1] denote a Participatory Configuration associated with realization object r over interval [t0,t1]. Πr is structured rather than scalar. It may preserve Subjects, Act Forms, RO/FP/AD relations, authority position, timing, revisability, and other pre-specified institutional attributes.
Πr[t0,t1] → Crj [consequence class j; linkage grade ℓ]
Crj is observed consequence class j and ℓ is the evidentiary linkage grade attached to that claim. Throughout Sections 9.8-9.9, → denotes only a documented or represented partial transition/sequence; it does not encode probability, causal effect, necessity, or sufficiency. The symbol ⇏ means that the transition is not entailed.
Πr ⇏ Cr
A second transition may occur when a Participatory Consequence changes an already-defined condition relevant to realization:
Crj → X′r
Chapter 4’s published realization architecture defines realization as:
Rg(t) = μt(P(t), Av(t), Exg(t), Zd(t))
Participation is not inserted as another generic argument. Where evidence supports the pathway, a consequence may alter an inherited realization condition represented within the Chapter 4 architecture. Frame One principally exercises institutional, infrastructural, exposure/recourse, and domain-condition pathways; it does not identify a generic Participation-to-P(t) transition.
Cr ⇏ ΔRg
The minimal representation is therefore:
Πr → Cr → X′r → R′g
Every arrow may fail. The expression Cr ⇏ ΔRg means that a consequence does not automatically produce realization change, not that consequences are never realization-relevant. Where a consequence alters X′r and the pathway continues, realization may change; F1-025, F1-044, and F1-048 are the observed examples. This is a typed partial-transition representation, not a pipeline.
9.8.2 Formal nonclaims
NewParticipationScalar = NotWarranted
DirectPiArgumentInMu = NotIdentified
Frame One does not justify a universal scalar Participation score p = σ(Π), where p would compress a structured Participatory Configuration into one quantity. Nor does it justify a common additive coefficient βΠ, a universal Act Form or RO effect, a Participation-specific threshold coefficient, or a monotone relation in which more Participation necessarily means more, less, faster, slower, or better realization. The empirical paths are too heterogeneous in sign, route, authority structure, and stopping point.
9.8.3 Adjudication and contraction conditions
For a domain-level Responsive Floor classification, the minimum conditions are:
1. an outcome-independent frame and comparator;
2. determinate participation states;
3. pre-specified and adequate precision;
4. sufficient comparability, stratification, or modeling of major realization conditions;
5. evidence that Participatory Configuration adds explanatory value beyond those conditions; and
6. explicit separation of documentary linkage from counterfactual causal identification.
If these conditions fail, the mandatory result is NotAdjudicated rather than automatic NotSupported. The same contraction discipline applies elsewhere: the Grammar cannot be promoted to a validated instrument while reproducibility is unresolved; Role Plasticity cannot be adjudicated without its automation antecedent; readiness cannot be promoted when the observation system does not measure it; and repeated association sequences cannot be promoted into an identified causal law.
9.8.4 Unresolved burdens
Relational Grammar: discriminant/nonredundancy evidence observed in the executed coding, conditional on faithful application of the instrument; intersubjective reproducibility cannot be reconstructed from the surviving archive, so instrument validation remains unresolved and requires prospective independent replication.
Full 3M validation remains unresolved because Manifestness and Materiality were not frozen confirmatory fields. Proposition A’s executed discrimination result therefore applies to RO/FP/AD against Act Form and ordinary descriptors, not to the integrated 3M Taxonomy.
Proposition B remains Untested in Frame One because the required human-task automation antecedent is absent.
Proposition B’s unresolved status also limits what Frame One can say about the Reciprocal Theory of Participation. The frozen design operationalized human Participatory Subjects and did not independently code the roles performed by human and AI actors within shared task sequences. It therefore cannot adjudicate reciprocal Participatory Role Plasticity: whether functions migrate or redistribute between human and AI actors over time, or whether one side’s acts alter the subsequent role or action space of the other. A prospective design would need to preserve the distinction between human Participatory Subject and actor role while separately observing human and AI functions, their redistribution, and their mutual conditioning. The broader 2026 AI-mathematics participatory field provides a motivating illustration of that design problem, not evidence that can be imported into Frame One.
Contrast B remains a pre-specified contrast not exercised because the required target-controlled comparator cell is absent. This is neither a null result nor a failed pre-registration.
Participatory Readiness remains measurement-limited. Temporal maturation remains prospective. The aggregate Responsive Floor remains NotAdjudicated. Port Washington and Lancaster do not identify the counterfactual contribution of Participation. Section 9.7’s ongoing 2026 transformation likewise remains a prospective generalization rather than a frozen empirical result.
measurement ≠ operability ≠ scope ≠ causal identification ≠ temporal generalization
The formal lesson is deliberately modest: Participation is structurally representable; its transitions are empirically contingent; and no universal Participation term is identified.
9.9 What Survives Empirical Contact
The purpose of empirical contact is not to make every conceptual proposition come out true. A stronger framework should survive a less convenient outcome: some distinctions prove useful, some propositions contract, some remain untested, some pathways appear without an aggregate effect, and some mathematically attractive extensions turn out not to be warranted.
That is what happened here.
9.9.1 Table 9.3 – What Survives Empirical Contact
| Claim or apparatus | Empirical result | Status after Chapter 9 |
| 3M Taxonomy | Useful descriptive representation; Act Form carries Modality while Manifestness and Materiality were added later | Empirically informed; full integrated validation unresolved |
| Proposition A – Conditional Non-Fungibility | Same or similar Act Forms preserve different RO, FP, and AD states in executed coding | Evidence: discrimination/nonredundancy observed, conditional on faithful application. Instrument: reproducibility unreconstructable; validation unresolved pending prospective replication. |
| Proposition B – Role Plasticity | Required automation antecedent not instantiated | Untested in Frame One |
| Act → Consequence → Realization separation | Bounded Null, information-only, procedure-only, unresolved, and realization-positive cases occur | Strongly exercised / operationally useful |
| Proposition C / Responsive Floor | Port Washington L3 documented non-veto sequence; Lancaster weaker L2; aggregate comparator sparse and heterogeneous | Documented non-veto association sequence observed; aggregate Responsive Floor NotAdjudicated |
| Participatory Readiness | 0/11 mature cases Present; 11/11 unresolved | Measurement-limited |
| 2026 resistance-frequency hypothesis | No positive calendar-2026 frequency effect identified | Not supported by Frame One |
| Universal Participation scalar or direct realization-mapping term | Heterogeneous signs, routes, stopping points, authority structures; no common coefficient | Not warranted / not identified |
9.9.2 What strengthened, and what contracted
Act ≠ Consequence ≠ Realization Change
A qualifying act can terminate in a bounded Null. An informational consequence can occur without procedural change. A procedural consequence can occur without realization change. Realization can be conditioned without project termination and can also be enabled through supportive authoritative action. These are observed empirical positions, not merely conceptual distinctions.
participation ≠ voice volume
recognition ≠ uptake
uptake ≠ realization
realization consequence ≠ success as defined by the participant
Three contractions are equally important. Conditional Non-Fungibility survives at the discrimination level but not as a fully validated instrument. Proposition C contracts from a possible domain-level relationship to a documented non-veto association sequence. And the simple temporal claim that resistance became more frequent in 2026 is not supported by Frame One; Section 9.7 instead motivates a prospective configuration hypothesis.
9.9.3 What remains unresolved, untested, or unexercised
The remaining burdens do not have the same status. Proposition B is untested because its antecedent is absent. Contrast B is unexercised because its comparator cell is absent. Participatory Readiness is measurement-limited. Aggregate Responsive Floor classification is NotAdjudicated because the comparator is sparse and the realization conditions heterogeneous. Causal contribution remains unidentified. Temporal institutionalization remains prospective.
unsupported ≠ untested ≠ unexercised ≠ measurement-limited ≠ not adjudicated
Preserving those statuses is part of the empirical contribution. A framework that treats every absence of confirmation as the same kind of failure would erase information produced by the design.
9.9.4 From empirical contact to epistemic discipline
The formal analysis reaches the same boundary. Frame One does not justify a universal scalar Participation score p = σ(Π), a direct fifth argument in μt, or a common additive coefficient βΠ. What survives is narrower:
Π → C → X’ → R’
with each transition potentially failing and → carrying the same non-causal, partial-transition meaning defined in Section 9.8.1.
Participation therefore survives empirical contact neither as a universal causal lever nor as an undifferentiated measure of human voice. It survives as a structured, partially observable process whose empirical status depends on where evidence enters and where inference must stop.
Frame One does not estimate the full reciprocal human-AI configuration. Chapter 8’s Reciprocal Theory of Participation therefore remains conceptually admissible and prospectively testable here, not a result inferred from this frozen empirical frame; reciprocal Participatory Role Plasticity likewise remains prospective rather than demonstrated by Frame One.
The chapter has repeatedly encountered the same epistemic limits: observability, missingness, censoring, archival provenance, reliability, comparator availability, documentary attribution, and the distinction between an observed pathway and an identified causal effect. These are not peculiarities of Participation. They are general problems for any attempt to measure a transformation whose realization remains incomplete and whose observable record is uneven.
That empirical posture is consistent with Ongoing, Unfinished Transformation, the book-wide architectural term established in Chapter 3 rather than a Chapter 9 temporal finding. Chapter 11 later takes up feedback and recursion explicitly; Chapter 9 does not pre-empt that burden.
represent → discriminate → consequence → realization → classification → formal limit
The early stages survive strongly. The later stages demand progressively more evidence. The framework becomes more useful, not less, when those burdens are allowed to remain unequal.
Chapter 10 therefore turns from Participation itself to Observation, Evidence, and Epistemic Discipline: how the Unified Gradual AGI Model should represent what is known, what is missing, what remains unresolved, and how evidentiary limits constrain the claims the model is permitted to make.
References and Contemporary Source Notes
Amodei, Dario. 2026. “We Must Pace the Frontier.” September 12, 2026. https://darioamodei.com/post/we-must-pace-the-frontier. Accessed September 14, 2026.
Annenberg Public Policy Center. 2026. “Opposition to Local Data Centers Rises Sharply, Annenberg Survey Finds.” August 11, 2026. https://www.annenbergpublicpolicycenter.org/opposition-to-local-data-centers-rises-sharply-annenberg-survey-finds/. Accessed September 14, 2026.
Associated Press. 2026. “Trump dismisses new AI guardrails, says there is a ‘SICK conspiracy’ against AI and data centers.” September 14, 2026. https://apnews.com/article/b85df16775ff7e9611a456b061a0e4b9. Accessed September 14, 2026.
Business Insider. 2026. “The hot new job in tech as AI apocalypse fears reach a tipping point.” September 14, 2026. https://www.businessinsider.com/what-is-embedded-evaluator-ai-apocalypse-dario-amodei-hire-2026-9. Accessed September 14, 2026.
W.H.L., GPT-5.6 Sol, and Claude Sonnet 5. 2026. “A Multi-Plateau Framework of the Unified Gradual AGI Model.” Chapter 3 of On Gradual AGI. Champaign Magazine, September 6, 2026. https://champaignmagazine.com/2026/09/06/a-multi-plateau-framework-of-the-unified-gradual-agi-model/. Accessed September 14, 2026.
W.H.L., GPT-5.6 Sol, and Claude Sonnet 5. 2026. “From Potential to Realization.” Chapter 4 of On Gradual AGI. Champaign Magazine, September 9, 2026. https://champaignmagazine.com/2026/09/09/from-potential-to-realization/. Accessed September 14, 2026.
W.H.L., GPT-5.6 Sol, and Claude Sonnet 5. 2026. “Participation: Subjects, Acts, and Consequences.” Chapter 8 of On Gradual AGI. Champaign Magazine, September 14, 2026. https://champaignmagazine.com/2026/09/14/participation-subjects-acts-and-consequences/. Accessed September 15, 2026.
W.H.L., GPT-5.6 Sol, and Claude Sonnet 5. 2026. “Gradual AGI as Participation: 3M Taxonomy, Relational Grammar, and Responsive Floor Test.” Gradual AGI Series #10, Publication Version v1.6.2, August 30, 2026. Champaign Magazine. https://champaignmagazine.com/2026/08/30/gradual-agi-as-participation-3m-taxonomy-relational-grammar-and-responsive-floor-test/
Engadget. 2026. Anna Washenko, “Meta Is ‘Pausing’ Employee Tracking Program After It Let The Whole Company See Sensitive Data.” June 22, 2026. https://www.engadget.com/2199458/meta-is-pausing-employee-tracking-program-after-it-let-the-whole-company-see-sensitive-data/. Accessed September 14, 2026.
Fast Company. 2026. Pavithra Mohan, “What happened when Meta tried to collect employee data.” September 11, 2026. https://www.fastcompany.com/91606241/what-happened-when-meta-tried-to-collect-employee-data. Accessed September 14, 2026.
Gardner Food and Agricultural Policy Survey. 2026. “Changing Public Perceptions of AI and Data Center Development.” farmdoc daily, September 14, 2026. https://farmdocdaily.illinois.edu/2026/09/changing-public-perceptions-of-ai-and-data-center-development-results-from-the-gardner-food-and-agricultural-policy-survey.html. Accessed September 14, 2026.
Microsoft, American Federation of Teachers, and United Federation of Teachers. 2026. “AFT, UFT and Microsoft announce ‘National AI Safety & Privacy Standard’ for schools to protect students, families and educators.” September 9, 2026. https://news.microsoft.com/source/2026/09/09/aft-uft-and-microsoft-announce-national-ai-safety-privacy-standard-for-schools-to-protect-students-families-and-educators/. Accessed September 14, 2026.
Reuters. 2026a. “Meta to start capturing employee mouse movements, keystrokes for AI training data.” April 21, 2026. https://www.reuters.com/sustainability/boards-policy-regulation/meta-start-capturing-employee-mouse-movements-keystrokes-ai-training-data-2026-04-21/. Accessed September 14, 2026.
Reuters. 2026b. “Meta scales back plan for internal mouse-tracking tech, citing staff concerns.” June 2, 2026. https://www.reuters.com/world/meta-scales-back-ai-mouse-clicks-tool-citing-employee-concerns-2026-06-02/. Accessed September 14, 2026.
Tao, Terence. 2026. “A Severe Misalignment of AI in Mathematics.” What’s New, September 11, 2026. https://terrytao.wordpress.com/2026/09/11/a-severe-misalignment-of-ai-in-mathematics/. Accessed September 14, 2026.
The Next Web. 2026. “‘Slop mathematics’: OpenAI walks away from a Caltech AI maths contest.” September 14, 2026. https://thenextweb.com/news/openai-withdraws-caltech-mathathon-slop-mathematics-fields-medallists. Accessed September 14, 2026.
Leave a Reply