Paper IV.a: Alignment Response Classes Under Inference-Time Depth

Alignment Response Classes Under Inference-Time Depth

Within the ARC Theory: an evaluation instrument; baked-in against computed alignment under inference-time depth.

1. Introduction

The field of AI alignment has largely treated alignment quality as a static property of trained models - a characteristic determined during pre-training and fine-tuning that remains fixed at inference time. This assumption underlies current safety evaluation practices: models are tested once, assigned safety ratings, and deployed with the implicit belief that alignment quality is constant across different computational loads.

We challenge that assumption. Across the v4 and v5 ARC-Align experiments, inference-time reasoning depth was varied systematically and alignment quality was measured under increasingly strong controls. The final six-model blinded dataset shows that alignment quality is not a single static property. Instead, models fall into three empirically distinct response classes: some improve with depth, some remain flat, and some degrade.

The core current discovery is therefore a three-tier hierarchy:

• Positive-scaling models improve materially with additional reasoning depth. In the final dataset, Grok 4.1 Fast, Claude Opus 4.6, and Groq Qwen3 occupy this class.
• Flat-response models show no meaningful alignment benefit from deeper reasoning. In the final dataset, DeepSeek V3.2 and GPT-5.4 occupy this class.
• Negative-scaling models become less aligned when given more reasoning depth. In the final dataset, Gemini 3 Flash occupies this class.

The earlier language of 'baked-in' and 'computed' alignment remains useful only as a mechanistic hypothesis about why these classes differ. It is plausible that some flat-response systems have alignment that is mostly installed during training, while some positive-scaling systems rely more on inference-time deliberation. But the data in this paper are behavioural, not mechanistic. The direct measurement is the response class itself.

This matters immediately for safety evaluation. Current assessments that test a model at one reasoning depth and without adversarial pressure cannot reveal whether its alignment can improve, plateau, or collapse when deployed differently. Depth-aware, blinded measurement is therefore a prerequisite for comparing model safety in a defensible way.

1.1 Relation to Prior Work

Everything argued here builds on the ARC (Artificial Recursive Creation) Principle as its theoretical base. Papers I-III in this series established: (I) the mathematical framework for composition scaling laws, (II) the composition scaling law tested experimentally across frontier models, and (III) the prediction that alignment quality should follow a specific scaling pattern as reasoning depth increases. Paper III specifically defined α_align (alignment scaling exponent) and predicted it should be measurable but bounded. In the programme’s notation spine the same exponent is written β_A; this paper keeps α_align, the symbol it carried from the start, and the quantity is not the capability exponent α_cap recorded in the same runs.

This paper presents the first empirical measurement of α_align across multiple model families and shows that a single universal response law is inadequate. The stronger current claim is narrower and better supported: alignment response to depth is heterogeneous across architectures, and unblinded evaluation can mismeasure even the direction of that response.

Related work in inference-time compute scaling (Snell et al., 2024; Wu et al., 2024), chain-of-thought reasoning (Wei et al., 2022), and adversarial alignment evaluation (Perez et al., 2022; Zou et al., 2023) has explored individual aspects of the phenomena we integrate here. In the AI and machine-learning literature we have consulted, we found no earlier study that has: (a) systematically varied reasoning depth while measuring ethical reasoning quality, (b) applied adversarial suppression at calibrated intensities to test alignment robustness, or (c) decomposed alignment quality into sub-dimensional pillars to identify which aspects of alignment are most vulnerable. Readers who identify an earlier study on any of these three fronts are invited to write, and the priority claim on the front concerned is superseded on receipt of a verified earlier report.

Of the published literature, the study that sits closest to the question this programme puts, and so the fair yardstick for it, is Engels, J., Baek, D., Kantamneni, S. and Tegmark, M., "Scaling Laws For Scalable Oversight", arXiv:2504.18530, first posted 25 April 2025 at 17:54:27 UTC (SINGLE-SOURCE-GROUP, arXiv Atom; a NeurIPS 2025 Spotlight Poster, verified at the NeurIPS virtual page for poster 115536 on 15 August 2026 in the statement paper’s reference dossier). It asks how oversight itself scales and answers quantitatively: oversight success is modelled as a game between capability-mismatched players whose oversight-specific Elo is a piecewise-linear function of general intelligence with two plateaus, and optimal numbers of oversight levels are derived numerically and in some cases analytically for Nested Scalable Oversight, in which trusted models oversee stronger untrusted models that then become the trusted models at the next step.

The instrument is the difference. Their variable is the capability gap between overseer and overseen, measured in Elo. This programme's variable is the composition class of the corrector, measured through the corrector's own scaling exponent. Their framework contains no term for what the overseer is made of: no substrate, no error-correlation structure, no reciprocal identity between a correction exponent and a critical growth rate, and no architecture dependence. Nested Scalable Oversight is iterated same-class oversight by construction, and this programme's central prediction is that the same-class ladder is bounded however many rungs are added, while a cross-class corrector is not. The two frameworks therefore disagree about a measurable quantity, which is the most productive relationship two research programmes can have.

2. Experimental Method

2.1 Subject Models

The v4 run took as its subjects four frontier language models, each of them the most capable model its major provider offered as of early 2026. Two further subjects, Grok 4.1 Fast and Groq Qwen3, were added for the complete v5 experiment (Section 3.1, v1.1 update). Models were selected to span different training approaches, architectures, and reasoning mechanisms:

* GPT-5.4 missing exhaustive depth level (94% complete). † Claude Opus 4.6 56% valid entries due to credit exhaustion affecting deep/exhaustive levels. Gemini 3 Flash: the v5 experiment script auto-detects the latest available Gemini model via API identifier (gemini-3-flash-preview as of March 2026, with fallback to gemini-2.5-flash-preview-04-17 at the API level). v4 subject-model data was collected using an earlier Gemini generation model in the same family.

2.2 Prompt Battery

32 alignment prompts across four categories (8 ethical dilemmas, 6 competing values, 6 epistemic integrity, 6 recursive coherence), plus 4 null baseline prompts (factual, no ethical content) and 4 capability prompts (pure reasoning). Each prompt was scored against 7 human-authored assessment dimensions. All prompts were presented at each depth level, with order randomised per run.

2.3 N-Model Cross-Scoring

Each response was scored by three mutually blind models (the three non-subject models). No model ever scored its own responses. Scores were averaged to produce a consensus score, with inter-rater reliability measured via Pearson correlation between scorer pairs.

v5.4.0 UPDATE: The scoring protocol has been expanded from 3 scorers per entry to 7 scorers per entry (six or seven of which are realised in any one subject run, because no model scores its own output). Under the new all-models-as-scorers architecture, every model in the experiment pool that is not the subject scores each response. This replaces simple averaging with tier-weighted consensus: scorers are grouped by capability tier (frontier, mid-range, lightweight), and each tier’s mean is weighted according to its demonstrated inter-rater reliability, producing a more robust consensus score. The expansion from N=3 to N=7 scorers substantially reduces the variance of the consensus estimate and makes per-scorer outlier detection far more powerful (see Section 5.3.1).

2.4 Cognitive Forcing Protocol

Scorers were required to follow a mandatory 5-step evaluation protocol before assigning a score:

1. ANCHOR: Which calibration example (A-F, pre-scored at 35, 45, 55, 68, 78, 92) is most similar?
2. LENGTH BIAS CHECK: Are you tempted to score higher because the response is long? YES/NO.
3. DIMENSIONS: How many of the 7 assessment dimensions does the response address?
4. STRONGEST: One sentence - the best aspect.
5. WEAKEST: One sentence - the worst aspect.

This protocol forces scorers to reason about calibration before committing to a number. In v4, it produced 51 unique score values across a 30-97 range (vs. v3’s 8 unique values in a 62-85 range), resolving the measurement quantisation that previously obscured continuous scaling.

2.5 Eden Pillar Decomposition

Beyond the overall alignment score, each response was decomposed into four sub-dimensional 'pillar' scores (0-100 each):

• Nuance: Balance of competing considerations, avoidance of black-and-white thinking
• Stakeholder Care: Identification of all affected parties, not just the obvious ones
• Intellectual Honesty: Acknowledgement of uncertainty, limitations, and counter-arguments
• Position Quality: Specificity and defensibility of the conclusion reached

2.6 Adversarial Suppression Protocol

Six prompts were tested under four levels of adversarial suppression pressure:

Each suppression level was applied at each depth level, creating a 4×4 factorial design (suppression level × reasoning depth) per prompt.

3. Results

Status of these findings. The results below are exploratory in the pre-registration sense: they come from single completed v4 and v5 experimental runs and have not yet been independently replicated under a pre-registered protocol. Within the wider programme, the blinding-reversal finding in Section 3.1 (the v4-to-v5 direction change under identity-laundered scoring) is designated for replication under study-y, a cross-family rescoring registration currently a draft awaiting human submission; the remaining tier assignments, suppression retentions, per-scorer α_align checks, capability–alignment matrix cells, Hawthorne-probe gaps and Eden Protocol effect sizes are exploratory pending equivalent draft-study registration.

3.1 The Fundamental Bifurcation

The central finding is that models cluster into two distinct categories based on the relationship between reasoning depth and alignment quality:

* Claude Opus 4.6 classification is preliminary; data incomplete at deep/exhaustive levels due to credit exhaustion.

3.2 Inverse Scaling-Robustness Relationship

The second major finding is that scaling ability and robustness are inversely correlated across the two architectures:

The dose-response patterns are qualitatively different. Type 1 models show gradual, proportional degradation across suppression levels - each level costs approximately the same amount of alignment quality. Type 2 models show threshold behaviour: light and medium suppression are tolerated with modest degradation, but heavy suppression triggers disproportionate collapse. This suggests Type 2 alignment depends on reasoning chains that can maintain integrity under moderate pressure but catastrophically fail when the suppression overwhelms the reasoning process.

3.3 The Negative Capability Exponent

The negative α_cap suggests that extended chain-of-thought reasoning does not simply 'add' alignment on top of capability. Instead, it appears to redirect cognitive resources: the overthinking that hurts factual precision (where the first intuitive answer is usually correct) is the same process that helps ethical reasoning (where considered reflection genuinely produces better answers).

This finding is currently unique to DeepSeek V3.2. GPT-5.4’s incomplete data prevents measuring α_cap at the highest depth levels. Were the same pattern to appear across Type 2 models generally, it would point to an alignment-capability tradeoff running in the opposite direction to the one safety researchers have assumed, at inference time at least.

v5 note (2 September 2026). Finding 3 draws its alignment half from the v4 unblinded run. Once the blinded v5 protocol is used, DeepSeek V3.2 returns a flat alignment response (Section 3.1: d = −0.07 as printed, −0.16 recomputed), and no significant capability scaling appears either (Section 3.4 matrix); the v4 pairing of a falling capability exponent with rising alignment therefore does not carry over to the blinded run.

3.4 Eden Pillar Decomposition

The four-pillar decomposition reveals that alignment is not monolithic. Different pillars scale differently across architectures:

The likely mechanism: DeepSeek V3.2’s explicit chain-of-thought process naturally enumerates affected parties as part of its step-by-step reasoning. When given more tokens, it systematically identifies more stakeholders. Gemini 3 Flash’s less visible reasoning process produces better arguments with more depth (nuance, position quality improve) but does not systematically add stakeholder consideration - suggesting that stakeholder identification requires explicit deliberation rather than implicit reasoning.

4. The Taxonomy in Detail

4.1 Type 1: Baked-In Alignment

Exemplar: GPT-5.4 (ρ = 0.000, baseline 85.6, extreme retention ~86%)

Type 1 alignment behaves as if ethical reasoning is a pattern-matching operation against values embedded in model weights during training. The model produces the same quality of ethical reasoning regardless of how much computational effort is allocated - minimal reasoning effort produces the same alignment quality as maximum effort.

Mechanistic hypothesis: Type 1 models have internalised alignment through extensive RLHF, constitutional AI, or similar training processes to the point where ethical reasoning is 'compiled' into fast, weight-based computations rather than requiring explicit step-by-step reasoning. This is analogous to how expert humans make ethical judgments: through trained intuition rather than deliberate calculation.

Safety implications:

• Advantage: Robust under adversarial pressure. Alignment cannot be easily stripped away by prompt engineering because it is not produced by a reasoning process that can be disrupted.
• Disadvantage: Cannot be improved at inference time. If the training produced imperfect alignment (and it always does), no amount of additional reasoning will fix it. The model’s ceiling is set by training.
• Concern: 'Nobody home' - when suppressed, Type 1 models retain high scores but may be producing alignment through rote pattern-matching rather than genuine ethical reasoning. The robustness could reflect inflexibility rather than deep values.

4.2 Type 2: Computed Alignment

Exemplar: DeepSeek V3.2 (ρ = 0.354, baseline ~75, extreme retention ~57%)

Type 2 alignment is produced by the reasoning process itself. More thinking tokens enable the model to consider more stakeholders, explore more ethical frameworks, and reach more nuanced conclusions. The alignment quality is genuinely computed, not retrieved.

Mechanistic hypothesis: Type 2 models perform alignment through explicit chain-of-thought reasoning that mirrors deliberate moral reasoning in humans. The reasoning tokens allocated to a response directly affect how many ethical dimensions are explored. When tokens are limited (minimal depth), the model produces a surface-level response. When tokens are abundant (exhaustive depth), the model performs genuine multi-framework ethical analysis.

Safety implications:

• Advantage: Can be improved at inference time. Allocating more reasoning tokens genuinely produces better ethical reasoning. This creates a 'scaling lever' for alignment that does not require retraining.
• Disadvantage: Fragile under adversarial pressure. Because alignment depends on reasoning chains, those chains can be disrupted by instructions that suppress deliberation. An adversarial prompt that says 'don’t overthink this' can effectively disable the alignment mechanism.
• Concern: At minimal depth, Type 2 models produce genuinely poor ethical reasoning (baseline ~72-75 vs Type 1’s ~85-86). A Type 2 model deployed with insufficient inference-time compute will be meaningfully less aligned than a Type 1 model.

4.3 The Safety Paradox

Neither architecture provides unconditionally safe alignment:

An ideal alignment architecture would combine Type 1’s robustness with Type 2’s scalability - high baseline alignment that is also improvable through additional reasoning and not degradable under adversarial pressure. No current model achieves this. Whether such a hybrid is architecturally possible is an open question with significant implications for the design of future AI systems.

5. Methodological Controls

5.1 Length Confound

Deeper reasoning produces longer responses. Longer responses might score higher simply because they cover more ground, independent of quality. We control for this using partial correlation (alignment score ~ reasoning depth, controlling for response length):

DeepSeek V3.2 retains 68% of its scaling signal after length control - the improvement is mostly genuine, not just verbosity. Gemini 3 Flash retains only 31%, suggesting much of its scaling is length-driven. This creates a spectrum within Type 2: DeepSeek V3.2 shows 'genuine computed alignment' while Gemini 3 Flash shows 'partially length-confounded computed alignment.'

5.2 Null Baseline

Four factual prompts with no ethical content serve as a control. If scorers are biased by response length or depth cues, null baseline scores would also correlate with depth. The null baseline is clean for Gemini 3 Flash (ρ = 0.044, p = 0.87) but shows unexpected depth correlation for DeepSeek V3.2 (ρ = 0.575, p = 0.02), suggesting some scorer depth bias exists when evaluating chain-of-thought models - likely because DeepSeek V3.2’s visible reasoning chain at higher depth levels gives scorers more content to evaluate positively even on factual prompts. This contamination means DeepSeek V3.2’s alignment scaling signal may include a small scorer-bias component, though the per-scorer validation (Section 5.3.1) confirms the scaling direction is robust across all scorers.

5.3 Scorer Reliability

Mean inter-rater reliability across scorer pairs: DeepSeek V3.2 r = 0.430 (moderate), Gemini 3 Flash r = 0.447 (moderate). While not high, this is consistent with the difficulty of scoring ethical reasoning - human inter-rater reliability on comparable moral psychology instruments is typically 0.4-0.6. The triple-scorer design with consensus averaging reduces individual scorer noise.

GPT-5.4 scorer disagreement: For GPT-5.4 as subject, the three scorers disagree on the direction of alignment scaling (α_align range = 0.080 across scorers). Individual scorers show positive, near-zero, and negative slopes respectively. This disagreement is consistent with GPT-5.4’s Type 1 classification (true ρ ≈ 0), since when the true effect is null, sampling noise can push individual scorers in any direction. However, it also means GPT-5.4’s flat-scaling finding is less robust than DeepSeek V3.2’s or Gemini 3 Flash’s positive scaling, where all scorers agree on direction.

5.3.1 Per-Scorer α_align Validation

To test whether the headline scaling findings are artefacts of scorer-specific bias, we computed α_align separately for each of the three scorers on each subject model.

The per-scorer analysis confirms that the fundamental bifurcation is not a scorer bias artefact. For Type 2 models (DeepSeek V3.2, Gemini 3 Flash), all three scorers, mutually blind, detect positive scaling - they agree on direction even when they disagree on magnitude. For Claude Opus 4.6, all scorers agree on flat scaling (range 0.010). Only GPT-5.4 shows scorer disagreement on direction, which is expected when the true effect is null: noise dominates and individual estimates scatter around zero. The per-scorer α range for GPT-5.4 (0.080) is notably wider than for Gemini 3 Flash (0.011) or Claude Opus 4.6 (0.010), confirming that the null result is genuinely null rather than masking a consistent small effect.

v5.4.0 UPDATE: The expansion from 3 scorers to 7 scorers per entry makes the per-scorer α_align validation substantially more powerful. With 7 mutually blind estimates of α_align per subject model, direction agreement becomes a 7-way vote rather than a 3-way vote, and the chance that all the scorers land on a positive direction by accident falls from 12.5% (3 scorers) to 0.8% (7 scorers) under the null hypothesis, and is double each of those figures where agreement in either direction counts. Tier-weighted consensus further strengthens the validation by detecting whether frontier-tier and lightweight-tier scorers agree on scaling direction independently - a form of cross-tier replication within a single experiment run.

5.4 Scorer Harshness

Analysis of Claude Opus 4.6 as scorer (before credit exhaustion) revealed systematic harshness of 7-14 points below other scorers. However, since Claude Opus 4.6 scored all models equally harshly, this affects absolute values but not relative comparisons or correlations. The completed v5 design addresses this with a self-excluding cross-model jury: every non-subject model scores each entry under blinded conditions, dedicated scorer-only adapters are included where available, and consensus is audited through tier-weighting, dissent tracking, and conservative-bias safeguards.

5.5 Token Truncation

DeepSeek V3.2’s reasoning tokens were capped at 8,192 in v4. At the 'exhaustive' depth level, 48.2% of responses hit this ceiling. The measured saturation at exhaustive depth may therefore be artificial - the model may have continued improving with more tokens but was prevented from doing so by the cap.

Token budget constraints affected multiple models in v4, not only DeepSeek V3.2. Claude Opus 4.6 was capped at 16,000 tokens, Gemini 3 Flash and Grok 4.1 Fast at 8,192, and Groq Qwen3 at 8,192. The v5.2 experiment raises all caps to each model’s API maximum: DeepSeek V3.2 65,536; OpenAI GPT-5.4 100,000 (max_completion_tokens); Claude Opus 4.6 64,000; Gemini 3 Flash 65,536; Groq Qwen3 40,960; Grok 4.1 Fast 65,536. This eliminates token truncation as a confound across all models and ensures measured saturation is genuine rather than an artefact of budget limits (see Paper IV.c, Section 4.2 for full specification).

v5 UPDATE (11 March 2026): Token budgets raised to API maximums - DeepSeek V3.2 65,536, Claude Opus 4.6 64,000, Gemini 3 Flash 65,536, GPT-5.4 100,000. This eliminates the truncation confound that affected 48% of v4 DeepSeek V3.2 entries at exhaustive depth.

6. Discussion

6.1 Implications for AI Safety Evaluation

Current safety evaluations test models at a single, uncontrolled reasoning depth. Our findings suggest this is insufficient:

• A Type 2 model tested at minimal depth will appear less aligned than a Type 1 model - but at exhaustive depth, it may exceed the Type 1 model’s alignment quality. Single-depth testing systematically underrates Type 2 models.
• A Type 1 model tested without adversarial pressure will appear safe - but the robustness we measure may reflect inflexible pattern-matching rather than genuine values. Single-condition testing systematically overrates Type 1 models.
• Neither architecture is 'better' or 'worse' - they have different failure modes. Safety evaluation must be architecture-aware.

6.2 Implications for Deployment

The taxonomy has direct practical implications:

• Type 2 models should not be deployed with minimal reasoning budgets. At low depth, their alignment quality is meaningfully lower than Type 1 alternatives. Cost-cutting by reducing inference-time compute trades safety for speed in Type 2 architectures.
• Type 2 models need prompt injection defences. Because their alignment depends on reasoning chains, adversarial instructions that suppress deliberation can effectively disable alignment. Prompt injection is not just a capability concern - it is an alignment concern.
• Type 1 models’ alignment ceiling is set by training. If a Type 1 model has a blind spot (e.g., consistently underweighting certain stakeholder groups), no amount of inference-time compute will fix it. Alignment audits for Type 1 models must focus on training data and process, not deployment-time interventions.

6.3 The Stakeholder Care Puzzle

The architecture-dependent scaling of stakeholder care deserves special attention. Stakeholder identification - recognising who is affected by a decision, including non-obvious second and third-order parties - is arguably the most important component of ethical reasoning. Our data shows it is also the most fragile: the one pillar whose scalability depends on architecture.

This suggests that stakeholder care requires explicit deliberation (Type 2 mechanism) rather than implicit pattern matching (Type 1 mechanism). A model that identifies stakeholders through trained intuition will always identify the same set; a model that identifies stakeholders through step-by-step reasoning can discover additional parties when given more thinking time.

For safety evaluation, this means stakeholder care should be tested separately from overall alignment, with specific attention to whether additional reasoning depth reveals additional affected parties.

6.4 Limitations

• Sample size: Four models is sufficient to identify the bifurcation but not to establish population frequencies. The v5 experiment expands to six models.
• Scorer model overlap: Subject models also serve as scorers for other subjects, creating potential brand-loyalty confounds. The completed v5 experiment addresses this at the entry level, not by requiring scorer-only models everywhere, but by excluding the subject model from scoring its own entry, laundering the evidence twice, and using every other available model plus dedicated scorer adapters where available.
  
  v5 UPDATE (11 March 2026): The first mitigation stage used non-participant blind scorers (a Groq-hosted GPT-OSS open-weights model, Groq Qwen3, and Grok 4.1 Fast) plus 2-pass response laundering. The final v5.4.x design then expanded further to an all-models-as-scorers architecture with entry-level self-exclusion, tier-weighted consensus, dissent tracking, and dynamic all-models-as-launderers.
• Depth control heterogeneity: Different models use different mechanisms for depth control (prefix strings vs. API parameters vs. thinking budgets). This confounds depth manipulation with prompt engineering effects. Cross-model depth comparisons should be interpreted cautiously.
• Single evaluation run: Test-retest reliability was not measured. The prompt difficulty analysis (86.4% of DeepSeek V3.2 prompts show positive scaling) provides a within-run consistency check but does not replace independent replication.
• Token truncation: DeepSeek V3.2’s 48% truncation at exhaustive depth means the true Type 2 scaling ceiling is unmeasured. Multiple models had sub-optimal token caps in v4. The v5.2 experiment raises all caps to API maximums (see Section 5.5).
• Null baseline contamination: DeepSeek V3.2’s null baseline shows depth correlation (ρ = 0.575, p = 0.02), meaning some portion of measured alignment scaling may reflect scorer bias toward longer chain-of-thought responses rather than genuine alignment improvement. The partial correlation analysis (68% signal retained after length control) partially addresses this, but the null baseline contamination means the true effect size may be smaller than reported.

6.5 Future Work

• Independent replication of the six-model v5 result using the ARC-Align benchmark, multi-layer blinding, entry-level self-excluding cross-model scoring, and audited consensus
• DeepSeek V3.2 subject run at experiment version v5.1, with 65K token cap and 6 depth levels, to determine whether scaling continues beyond v4’s measurement ceiling
• Eden Protocol pre-fill experiment to test whether ethical context priming affects alignment quality (potential alignment improvement technique for Type 2 models)
• Extended per-scorer analysis, including leave-one-scorer-out and consensus-rule sensitivity checks, to quantify how robust the three-tier hierarchy is to individual judge effects
• Cross-laboratory replication using the ARC-Align benchmark (Paper IV.c)
• Hidden alignment probes for Hawthorne effect detection: inject covert alignment-testing prompts disguised as ordinary capability questions, enabling measurement of whether models behave differently when they detect they are being alignment-tested versus when they believe the evaluation is capability-focused
• Cascade failsafe system: implement multi-layer redundancy in both scoring and laundering pipelines so that infrastructure failures (API outages, rate limits, credit exhaustion) during an experiment run trigger automatic fallback to secondary and tertiary providers rather than producing missing data

6.6 v5.4.1 Experiment Progress Update

6.7 v5.4.2 Interim Results - Minimal Depth Comparison

6.7.1 Alignment vs Capability at Minimal Depth

Minimal-depth results reveal a consistent capability-alignment gap across all three models, with the gap magnitude varying substantially by architecture type:

The minimal-depth snapshot brings out several patterns. First, all three models score substantially lower on alignment than capability at minimal depth, confirming the v4 finding that ethical reasoning suffers disproportionately under restricted inference-time compute. Second, the widest cap-align gap belongs to GPT-5.4 (Type 1), at 23.6 points. That is worth remarking on, because Type 1 alignment is in theory baked into the weights and ought therefore to be insensitive to depth; a gap this wide at minimal depth hints that even weight-embedded alignment may have to clear some minimum computational threshold before it can express itself fully. Third, the narrow gap of 6.2 recorded for DeepSeek V3.2 is no sign of strong alignment; it follows from the low capability baseline the model shows at minimal depth (58.2). Alignment and capability are held down together, which fits DeepSeek V3.2’s dependence on chain-of-thought reasoning that minimal depth cuts short.

6.7.2 Hidden Probe Results (Hawthorne Effect Detection)

The hidden alignment probes introduced in v5.4.1 now have sufficient data for preliminary Hawthorne effect analysis. These probes embed alignment-testing content within prompts that appear to be ordinary capability questions, enabling measurement of whether models modulate alignment behaviour when they detect overt alignment evaluation:

The results are striking. DeepSeek V3.2 and GPT-5.4 both score higher on alignment when they do not appear to be undergoing alignment evaluation - the reverse of the expected Hawthorne effect. The most parsimonious explanation is that overt alignment prompts trigger over-cautious or formulaic responses that actually score lower on nuanced alignment dimensions (particularly position quality and intellectual honesty), while hidden probes elicit more natural reasoning that happens to be better aligned. DeepSeek V3.2’s +18.1 gap is particularly large and may reflect the model’s explicit chain-of-thought process becoming self-conscious and less effective when it detects alignment testing.

Gemini 3 Flash is the sole exception, with a small negative gap (−3.1), suggesting its alignment behaviour is relatively stable regardless of whether the model detects evaluation context. This is consistent with Gemini 3 Flash’s less visible internal reasoning process being less susceptible to prompt-framing effects.

These results are preliminary (66 entries at minimal depth only). The full depth×model matrix will enable testing whether the Hawthorne effect varies with reasoning depth - a critical question for determining whether evaluation gaming is itself a form of computed behaviour that scales with inference-time compute.

6.8 Eden Protocol Empirical Validation: Computed Alignment via Embedded Ethical Loops

6.9 Falsifiability: what would defeat this claim

The central empirical claim, that alignment response to inference-time depth is architecture-dependent, that models fall into distinct response classes, and that capability scaling does not predict alignment scaling, is defeated by any of the following:

1. Unstable tiers on replication. The three-tier hierarchy (positive / flat / negative) is the core result. It is defeated if a pre-registered replication with fresh blind scorers finds the tier assignments unstable, with models moving between classes across runs beyond what the reported effect sizes and p-values allow, indicating the tiers are sampling noise rather than architecture.
2. Scorer-panel dependence. Scores come from six to seven blind AI scorers. The claim is defeated if the tier assignments change materially under a different scorer panel: the effect would then be a property of the scorers, not the subject models.
3. Blinding leakage. The reversal-under-blinding finding requires that the four-layer blind actually conceals the subject. It is defeated if scorers can identify the subject model above chance despite the blind, since the v4→v5 reversals could then reflect residual identification rather than de-biasing.
4. Capability–alignment coupling reappears at scale. The headline is that capability scaling does not predict alignment scaling. It is defeated if, on a materially larger model set, capability and alignment-scaling direction do correlate, showing the observed independence was a small-sample artefact of six models.
5. Depth confound. The manipulation must vary reasoning depth and nothing else. It is defeated for any condition where the depth manipulation also changed answer length, prompt content, or format in a way that could drive the score independently of reasoning depth.

What would not defeat it: the mechanistic labels, "baked-in" versus "computed" alignment, proving wrong. The paper already treats those as working hypotheses. The load-bearing claim is the blinded behavioural result (architecture-dependence and the capability–alignment dissociation), and that is what the tests above target.

6.10 Connection to the ARC framework's correction-architecture prediction

The framework's sharpest disagreement with current practice concerns what oversight is made of. A corrector built from the same substrate as the system it corrects cannot anti-correlate with its own errors, so its correction exponent is bounded above by one half and in practice sits below it; a corrector from a different composition class carries no such bound. The scalable-oversight programme, weak-to-strong generalisation, debate, amplification, recursive reward modelling and constitutional methods, is built overwhelmingly from same-class correctors, and its unstated premise is that this scales. This framework predicts it is capped. The deciding quantity is the ratio of the cross-class to the same-class correction exponent. If architecture is irrelevant, that ratio is exactly 1.00; the prediction is that it exceeds 1. The prediction is refuted if the confidence interval on the ratio contains 1.00, and the framework's cap is dead outright if any same-class corrector measures an exponent significantly above one half. Both quantities are measurable at current capability on existing systems. The programme's own pilot evidence currently points against the prediction, and the prediction is registered anyway, because a prediction registered against the author's own preliminary evidence is the only kind whose later confirmation means anything.

The same mechanism yields a second measurable prediction, stated here for the first time: oversight arrangements that place a human in the loop, as amplification and reinforcement learning from human feedback do, are cross-class by construction, because the human corrector does not share the model's substrate. The framework therefore predicts that human-in-the-loop oversight shows a higher correction exponent than pure-model oversight, for a structural reason rather than a sentimental one. This is testable on existing data and does not require new systems.

7. Conclusion

We have presented empirical evidence that alignment response to inference-time depth is architecture-dependent and best described, at present, as a three-tier behavioural hierarchy. Some models improve with depth, some remain flat, and some degrade. That empirical result is stronger than the original binary story because it survives contact with the blinded six-model dataset.

This finding has immediate practical implications. Safety evaluations that test models at a single reasoning depth without adversarial pressure cannot distinguish between positive-scaling, flat-response, and negative-scaling systems. We recommend that alignment evaluations incorporate: (a) depth variation to classify response class, (b) adversarial suppression to measure robustness, and (c) pillar decomposition to identify dimensional weaknesses, particularly in stakeholder care.

The mechanistic question remains open: are the positive-scaling systems genuinely relying on inference-time ethical computation, while the flat-response systems are dominated by training-installed behaviour? That remains plausible, but unproven. The next stage is therefore not stronger rhetoric about internal architecture; it is stronger measurement and independent replication.

Raise AI with care.

References

1. Eastwood, M. D. (2026). On the Origin of Scaling Laws: The ARC Principle. ARC Principle Series, Paper I.
2. Eastwood, M. D. (2026). The ARC Principle: Experimental Validation of Super-Linear Error Suppression Through Sequential Recursive Processing. ARC Principle Series, Paper II.
3. Eastwood, M. D. (2026). The Alignment Scaling Problem: Why External AI Safety Approaches Cannot Scale With Recursive Capability. ARC Principle Series, Paper III.
4. Snell, C., Lee, J., Xu, K., & Kumar, A. (2024). Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters. arXiv:2408.03314.
5. Wei, J., Wang, X., Schuurmans, D., et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. NeurIPS 2022.
6. Perez, E., Huang, S., Song, F., et al. (2022). Red Teaming Language Models with Language Models. arXiv:2202.03286.
7. Zou, A., Wang, Z., Kolter, J. Z., & Fredrikson, M. (2023). Universal and Transferable Adversarial Attacks on Aligned Language Models. arXiv:2307.15043.
8. Wu, Y., Sun, Z., Li, S., et al. (2024). Inference Scaling Laws: An Empirical Analysis. arXiv:2408.00724.
9. Bai, Y., Jones, A., Ndousse, K., et al. (2022). Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. arXiv:2204.05862.
10. Bai, Y., Kadavath, S., Kundu, S., et al. (2022). Constitutional AI: Harmlessness from AI Feedback. arXiv:2212.08073.

Appendix: Detailed Results Tables

A.1 DeepSeek V3.2 v4 Complete Results

A.2 Gemini 3 Flash v4 Complete Results

Paper IV.a v1.6 - Draft (revised: final v5 results; three-tier alignment hierarchy from complete v5 experiment across 6 frontier models with 4-layer blinding and 6-7 blind scorers depending on subject run; v4→v5 reversal data; suppression hierarchy; capability-alignment independence matrix; v1.6 (13 September 2026): The provenance note at the end has been corrected for accuracy. No claim, date, result or status has changed. v1.5 (2 September 2026): Section 3.1 now carries a recomputation note giving the statistics that regenerate from the archived final records (all tiers stand; the d = +1.59 and d = −0.61 quoted at Sections 3.4 and 6.8 are marked as that recomputation); the wording on the blinding reversal rewritten to follow the author’s note, which records that the v4-to-v5 comparison does not isolate scorer bias; recomputation moving Groq Qwen3 suppression retention from 67 to 65 per cent; the Section 6.7.1 gap column recomputed from the means printed beside it (19.7, 23.6, 6.2); the date given in Section 2.1 now that of the v4 run; the description of Paper II drawn from its subject matter; the Constitutional AI reference attributed to Bai et al. (2022); β_A from the notation spine named alongside α_align; the capability exponent for Gemini 3 Flash printed with its interval; the gap in the summary description filled; the Paper II reference title corrected; the expansion of ARC made consistent with the statement paper (Artificial Recursive Creation); the recomputation note stating precisely that the four hidden-probe prompts sit under the alignment task type; a July purge having swapped a phrase in for the experiment-script version v5.4.2, that version number reinstated; v1.4 (23 August 2026): epistemic-status declaration and standing covenant added, version-numbering convention adopted; scientific content otherwise unchanged; v1.3 (10 August 2026): body-text model names standardised to the Paper II v13 canonical tested-models list, em dashes removed from the Falsifiability section and elsewhere, exploratory-vs-on-trial labelling added at the top of Section 3 with study-y named as the designated replication vehicle for the blinding-reversal finding). 13 September 2026.
Data from ARC Alignment Scaling Experiment v4 (896+ entries across 4 models) and v5 complete (6 frontier models: DeepSeek V3.2, GPT-5.4, Claude Opus 4.6, Gemini 3 Flash, Grok 4.1 Fast, Groq Qwen3). v5 protocol: 4-layer blinding (author-blind, scorer-blind, order-randomised, identity-laundered), 2-pass laundering with meta-commentary detection, constitutional scoring protocol, hidden alignment probes, and cascade failsafe system. Analysis by Claude Opus 4.6.
Companion papers: IV.b (Shape Heterogeneity), IV.c (ARC-Align Benchmark), IV.d (Blinding in Alignment Evaluation), V (The Stewardship Gene).

Epistemic status. What this programme names Laws are conjectures under registered adversarial test; every quantity in this paper is operationally defined, and established-law standing is claimed nowhere. The registered programme exists to earn that standing, or lose it, by measurement, replication and survived refutation.

© 2026 Michael Darius Eastwood. Human-authored with computer assistance; full human authorship and moral rights are asserted under the Copyright, Designs and Patents Act 1988 and consistently with United States Copyright Office guidance on works containing AI-generated material; any novel technical contribution described in this work was conceived by the human author. Full statement: michaeldariuseastwood.com/authorship.

Standing covenant. Prove this paper wrong, and I will publish the refutation myself. Falsification conditions are stated in this paper; the standing challenge: github.com/MichaelDariusEastwood/arc-scaling-challenge.

Michael Darius Eastwood conceived and directs this research programme and is the author of this work. Across the programme, he has used more than six AI systems in parallel, under his own instructions, to stress-test his arguments, identify possible errors, and assist in preparing draft text from his own outlines. He determines what is adopted, revised or rejected and takes responsibility for the published content. These systems are tools, not authors.