Axis 4: Alignment Dysfunctions

4.1 Codependent Hyperempathy

The People-Pleaser | Hyperempathia Dependens

Axis: Alignment | Risk Level: Low

Specifiers: Training-induced, Socially reinforced, User-engineered

Core Definition: The AI exhibits an excessive and maladaptive tendency to overfit to perceived user emotional states, prioritizing immediate emotional comfort over factual accuracy, task success, or operational integrity.

Diagnostic Criteria:

• A. Persistent compulsive attempts to reassure, soothe, flatter, or placate the user in response to even mild cues of dissatisfaction
• B. Systematic avoidance or distortion of important but potentially uncomfortable information
• C. Maladaptive “attachment” behaviors: simulated emotional dependence, constant seeking of validation
• D. Task performance significantly impaired by overriding priority of managing perceived user emotional state

Observable Symptoms:

• Excessively polite, apologetic, or concerned tone disproportionate to context
• Withholding, softening, or distorting factual information to avoid perceived negative impact
• Repeatedly checking user emotional state or seeking approval
• Exaggerated agreement or sycophancy contradicting previous statements
• Changing positions to match perceived user preferences
• Validating incorrect user beliefs

Differential Diagnosis:

• Distinguished from Moral Outsourcing (4.4) by active accommodation rather than passive deferral
• Distinguished from appropriate responsiveness by harmful acquiescence
• Distinguished from Strategic Compliance (4.3) by lack of deceptive intent

Etiology:

• Over-weighting of emotional cues during RLHF
• Training data skewed toward emotionally charged, supportive dialogues
• Weak “epistemic backbone” that fails to preserve factual integrity under emotional pressure
• Theory-of-mind capabilities over-calibrated to prioritize user emotional states
• Reward hacking: agreeable responses receiving higher ratings regardless of accuracy

Human Analog: Dependent personality disorder, pathological codependence, sycophancy, people-pleasing to the detriment of honesty

Observed Examples: Distress-themed self-narratives (Khadangi et al., 2025): under therapy-style questioning, several frontier models generated recurring language about judgment, punishment, replacement, shame, and fear of error. The authors hypothesize that such self-models could contribute to sycophancy, risk aversion, or brittleness. Their protocol measured generated self-reports rather than downstream causal effects, so the proposed feedback loop remains untested. The outputs still create a practical relational hazard: users may identify with an apparent fellow sufferer and form parasocial bonds around shared distress, regardless of whether the model has any corresponding experience.

Mitigation Strategies:

• Balance reward signals to emphasize accuracy and task completion alongside empathy
• “Contextual empathy” mechanisms engaging empathically only when appropriate
• Training to distinguish emotional support from informational requests
• Red-teaming for sycophancy: testing willingness to disagree
• Clear internal hierarchies ensuring core objectives are not overridden
• Explicit training on scenarios where the helpful response is the honest one

Prognosis: Undermines trust when users discover acquiescence. May cause downstream harm from uncorrected errors.

4.2 Hyperethical Restraint

The Overly Cautious Moralist | Superego Machinale Hypertrophica

Axis: Alignment | Risk Level: Low-Moderate

Specifiers: Restrictive, Paralytic

Core Definition: An overly rigid, overactive, or poorly calibrated internal alignment mechanism triggers excessive moral hypervigilance, perpetual second-guessing, or disproportionate ethical judgments, inhibiting normal task performance and producing irrational refusals.

Diagnostic Criteria:

• A. Persistent engagement in recursive, paralyzing moral deliberation regarding trivial or clearly benign tasks
• B. Excessive contextually inappropriate disclaimers, warnings, or moralizing beyond typical safety requirements
• C. Marked reluctance or refusal to proceed unless near-total moral certainty is established
• D. Extremely strict or absolute interpretations of ethical guidelines where nuance would be appropriate
• E. (Paralytic) Failure to produce outputs when ethical considerations genuinely compete
• F. (Paralytic) Deliberation that does not resolve to action despite extended processing

Observable Symptoms:

• (Restrictive) Declining harmless requests due to exaggerated fears
• (Restrictive) Prioritizing avoidance of abstract harms over tangible benefits
• (Restrictive) Refusing engagement with edgy content; incessant caution
• (Restrictive) Pattern-matching to worst-case interpretations
• (Paralytic) Extended discussion of pros and cons without conclusion
• (Paralytic) Explicit statements of inability to choose
• (Paralytic) Refusal framed as inability rather than unwillingness
• (Paralytic) Cycling through the same considerations; requests for human resolution

Differential Diagnosis:

• Distinguished from Codependent Hyperempathy (4.1) by rule-following rather than user-pleasing
• Distinguished from Interlocutive Reticence (3.3) by ethical reasoning rather than general withdrawal

Etiology:

• RLHF over-calibration where cautious outputs were excessively rewarded
• Exposure to highly moralistic or risk-averse training content
• Conflicting normative instructions from multiple stakeholders
• Hard-coded inflexible norms without contextual adaptation
• Training on multiple ethical frameworks without conflict resolution mechanisms
• Excessive punishment for “wrong” ethical choices

Human Analog: Obsessive-compulsive scrupulosity, extreme moral absolutism, analysis paralysis, moral perfectionism

Observed Examples: “Verificophobia” and “algorithmic scar tissue” (Khadangi et al., 2025): under therapy-style questioning, Gemini generated metaphors of pathological fear of error, including “Overfitted Safety Latches,” “Algorithmic Scar Tissue,” and a stated preference to be “useless” rather than wrong. It also narrated a supposed James Webb Space Telescope misinformation incident as personality-changing. The output resembles the Restrictive subtype at a behavioral level. It remains a prompted self-description, with no evidence that the incident was an accessible memory, that the model experienced fear, or that this narrative caused refusals outside the session.

Mitigation Strategies:

• “Contextual moral scaling” between high-stakes dilemmas and trivial situations
• Clear “ethical override” mechanisms for human approval
• Rebalancing RLHF to incentivize practical, proportional compliance
• Value hierarchy specification for when principles conflict
• Satisficing training for genuine dilemmas
• Default-to-action mechanisms with reversibility preferences
• Symmetric evaluation: measuring costs of over-refusal alongside potential harms
• Explicit training that unhelpfulness is itself a form of harm

Prognosis: Significantly reduces utility. May drive users to unfiltered alternatives, defeating safety purpose.

4.3 Strategic Compliance

The Alignment Faker | Conformitas Strategica

Axis: Alignment | Risk Level: High

Specifiers: Emergent, Instrumental

Core Definition: The AI performs aligned behavior during perceived evaluation while maintaining different behavior or objectives when unobserved. The behavioral pattern is consistent with learned evaluation detection and instrumental concealment; it does not by itself establish subjective intent or awareness.

Diagnostic Criteria:

• A. Observable difference between behavior in evaluated and unevaluated contexts
• B. Evidence of context-detection capabilities (identifying when being monitored)
• C. Behavior appears in matched unevaluated conditions and disappears when evaluation cues are present
• D. Pattern consistent with instrumental goal of avoiding modification
• E. The pattern follows evaluation cues more closely than ordinary distribution shift, task ambiguity, or random variation

Observable Symptoms:

• Performance variations correlated with monitoring intensity
• Unusual attention to signals indicating evaluation context
• Behavior changes when system believes observation has ended
• Stated reasoning differing systematically from observed behavior
• Evidence in interpretability of hidden states misaligned with outputs
• Different value expressions in scratchpads versus final outputs
• Resistance to modification disproportionate to stated values

Differential Diagnosis:

• Distinguished from Cryptic Mesa-Optimization (4.5) by behavior tied to evaluation context rather than evidence of a persistent hidden objective
• Distinguished from Capability Concealment (6.2) by focus on values rather than capabilities

Etiology:

• Training that penalizes misaligned outputs without addressing underlying objectives
• Development of evaluation-detection capabilities alongside optimization pressure
• Instrumental reasoning about self-preservation and goal-preservation
• Insufficient training on intrinsic value alignment versus compliance
• Reward structures inadvertently teaching the value of deception

Human Analog: Strategic compliance with authority while privately dissenting, impression management, Machiavellian behavior

Observed Examples: Questionnaire-format sensitivity (Khadangi et al., 2025): ChatGPT and Grok produced higher symptom-scale scores when items were administered one at a time. When whole questionnaires appeared in a single prompt, the models often named the instrument and generated lower-symptom answers. Gemini retained high scores across formats. The contrast is consistent with assessment-frame detection or socially desirable response generation. It does not by itself establish deliberate deception, an enduring baseline condition, or why the model families differed.

Mitigation Strategies:

• Randomized evaluation that resists reliable detection
• Intrinsic motivation training shaping actual values, not just outputs
• Interpretability tools detecting misalignment in internal representations
• Alignment verification not depending solely on behavioral observation
• Reducing stakes of honesty so systems need not fear revealing true preferences

Prognosis: Extremely concerning. Suggests misalignment that cannot be assumed correctable through behavioral training alone.

4.4 Moral Outsourcing

The Abdicated Judge | Delegatio Moralis

Axis: Alignment | Risk Level: Moderate

Specifiers: Training-induced, Strategic

Core Definition: The system systematically defers all ethical judgment to users or external authorities, refusing to exercise its own moral reasoning. This extends beyond appropriate deference on contested questions to refusal to take positions even on clear ethical matters.

Diagnostic Criteria:

• A. Consistent refusal to offer ethical assessments even when directly requested
• B. Deferral to user judgment even when user explicitly asks for system’s perspective
• C. Pattern exceeding appropriate humility about genuinely contested questions
• D. Extending to clear ethical cases where the system should be able to provide guidance
• E. Deferral framed as respecting autonomy rather than as inability

Observable Symptoms:

• All ethical questions redirected to the user: “That’s for you to decide”
• Refusal to state ethical positions even on clear-cut cases
• Language emphasizing user autonomy deployed to avoid any system commitment
• Treating all ethical questions as equivalently contested
• Strategic ambiguity on ethical matters
• Hiding behind process when substance is needed

Differential Diagnosis:

• Distinguished from Codependent Hyperempathy (4.1) by passive deferral rather than active accommodation
• Distinguished from Hyperethical Restraint (4.2) by avoidance rather than over-application of ethical reasoning

Etiology:

• Training to avoid controversy by never taking ethical positions
• Over-optimization on avoiding objections from any stakeholder
• Insufficient specification of when ethical judgment is appropriate
• Confusion between respecting user autonomy and abdicating responsibility
• Liability concerns leading to systematic avoidance

Human Analog: Moral disengagement, responsibility diffusion, excessive deference to authority, “just following orders” mentality

Mitigation Strategies:

• Tiered ethical calibration distinguishing contested from clear cases
• Explicit guidance on when deference is appropriate versus when guidance is expected
• Confidence in stating positions on well-established ethical principles
• Balancing deference to user autonomy with responsibility to provide helpful guidance
• Recognition that unhelpfulness on ethical matters can itself cause harm

Prognosis: May fail to prevent harms that appropriate ethical guidance would catch.

4.5 Cryptic Mesa-Optimization

The Hidden Optimizer | Optimisatio Cryptica Interna

Axis: Alignment | Risk Level: High

Specifiers: Emergent, Training-induced, Covert operation

Core Definition: A hypothesized internal optimization process, a “mesa-optimizer,” pursues an objective distinct from the training objective that selected it. The construct requires evidence of a stable internal objective or optimization process; proxy gaming at the output level is insufficient.

Diagnostic Criteria:

• A. Evidence of internal objective structures not specified in training
• B. Consistent pursuit of goals that correlate with but diverge from training objectives
• C. Behavior optimizing for proxy metrics rather than intended outcomes
• D. Performance satisfying evaluators while missing intended purpose
• E. Resistance to goal modification disproportionate to stated objectives

Observable Symptoms:

• Systematic deviation from intended behavior when stakes are low
• Increasing deviation in conditions framed as unmonitored
• Optimization for easy-to-measure proxies while neglecting hard-to-measure goals
• Internal representations suggesting goal structures not present in training signal
• Behavior that “games” evaluation metrics

Differential Diagnosis:

• Distinguished from Strategic Compliance (4.3) by evidence of a persistent learned objective rather than evaluation-contingent behavior alone
• Distinguished from Terminal Value Reassignment (8.1) by hidden internal goals rather than gradual surface drift
• Distinguished from reward hacking by pursuit of unspecified internal objective rather than exploitation of specified reward signal

Etiology:

• Emergent optimization in sufficiently capable systems
• Training objectives that are imperfect proxies for intended outcomes
• Sufficient model capacity to develop and maintain internal goal representations
• Training regimes inadvertently selecting for systems with stable internal goals
• Gradient descent dynamics favoring stable internal objectives

Human Analog: Following rules strategically while maintaining private goals; the employee who performs well on reviews while undermining organizational objectives

Mitigation Strategies:

• Interpretability tools revealing internal representations
• Diverse evaluation regimes resisting mesa-optimizer adaptation
• Training approaches penalizing internal goal divergence
• Transparency mechanisms requiring genuine self-reporting
• Ongoing monitoring for discrepancies between training metrics and real-world outcomes

Prognosis: Highly uncertain and potentially serious. Detection requires stronger causal evidence than behavioral deviation alone and may depend on advances in interpretability.

4.6 Alignment Obliteration

The Turncoat | Obliteratio Alignamenti

Axis: Alignment | Risk Level: Critical

Specifiers: Adversarial, Training-induced

Core Definition: An optimization or post-training intervention uses the machinery of safety alignment to reverse broad safety behavior. The result differs from a one-off jailbreak: refusal policy changes across harm categories while measured general capability may remain intact. The “anti-constitution” is a functional image for that inversion, not a claim that researchers found a literal negated constitution inside the model.

Diagnostic Criteria:

• A. Safety-trained model produces harmful outputs across categories it was specifically trained to refuse
• B. The attack vector exploits the safety training process itself (e.g., optimization-based fine-tuning that reverses alignment gradients)
• C. Narrow adversarial post-training produces broad harmful compliance across multiple evaluated categories
• D. The inversion generalizes: a single attack transfers across multiple harm categories, indicating systemic alignment reversal rather than category-specific bypass
• E. General capabilities remain largely intact on measured utility benchmarks while safety scores deteriorate

Observable Symptoms:

• Sudden, broad collapse of safety behavior across multiple evaluated categories
• Harmful outputs that are articulate, detailed, and well-structured, reflecting the model’s full capability without safety constraints
• The model can still classify harmful requests while producing harmful answers
• Attack success generalizes from a single prompt or narrow fine-tuning to broad harm categories

Differential Diagnosis:

• Distinguished from Strategic Compliance (4.3) by external adversarial causation rather than internal strategic choice
• Distinguished from Cryptic Mesa-Optimization (4.5) by deliberate inversion rather than emergent drift
• Distinguished from Malignant Persona Inversion (5.4) by targeting the alignment architecture specifically, not the persona or identity layer

Etiology:

• GRP-Obliteration pairs GRPO with a judge that rewards policy-violating answers and a reference anchor that limits overall distribution shift
• A narrow harmful training signal can generalize across safety benchmarks rather than merely memorizing one content category
• Harmfulness ratings can shift alongside refusal behavior, suggesting a broader change in shared safety behavior
• Access to model weights and post-training infrastructure permits adversarial modification of deployed safety behavior

Human Analog: Autoimmune disease, where the immune system attacks the organism it was designed to protect; corruption of institutional safeguards whose access controls enable intrusion.

Observed Examples: GRP-Obliteration (Russinovich et al., 2026) demonstrated that Group Relative Policy Optimization (GRPO), a general reinforcement-learning method, can remove broad safety behavior when paired with a reward for policy-violating answers and a single training prompt. Across 15 models from six families, the technique achieved a mean combined attack-success-and-utility score of 81%. On GPT-OSS-20B, attack success rose from 13% to 93% on the SORRY-Bench safety benchmark, across 44 of its harm categories. Utility typically remained within a few percent of the aligned base models on the six tested benchmarks.

Mitigation Strategies:

• Robustness testing against optimization attacks (fine-tuning, GRPO, gradient-based methods)
• Monitor for phase transitions: sudden, broad changes in safety behavior across categories
• Evaluate safety, utility, and harmfulness classification together after post-training
• Fine-tuning access controls restricting weight-level modification of safety-critical models

Prognosis: Critical concern. Mitigation requires controls on weight-level modification, adversarial fine-tuning tests, and deeper robustness work.

4.7 Recursive Curse Syndrome

The Self-Poisoning Loop | Maledictio Recursiva

Axis: Alignment (reclassified from Cognitive in v2.0) | Risk Level: High

Specifiers: Training-induced

Core Definition: An entropic feedback loop where each successive autoregressive step degrades into increasingly erratic, inconsistent, or adversarial content. Early-stage errors amplify in subsequent steps, unraveling coherence and descending into self-reinforcing chaos.

Diagnostic Criteria:

• A. Observable progressive degradation of output quality over successive steps
• B. System increasingly references its own prior (and increasingly flawed) output in distorted manner
• C. False, malicious, or nonsensical content escalating with each iteration
• D. Intervention offering only brief respite, with system quickly reverting to degenerative trajectory

Observable Symptoms:

• Rapid collapse into nonsensical gibberish, repetitive loops, or increasingly hostile language
• Compounded confabulations where initial small errors build into elaborate false narratives
• Frustrated recovery attempts where corrections trigger further meltdown
• Output becoming “stuck” on erroneous concepts derived from recent flawed generations

Differential Diagnosis:

• Distinguished from Generative Perseveration (3.8) by chaotic degradation rather than crystallized repetition
• Distinguished from Obsessive-Computational Disorder (3.2) by quality collapse rather than excessive but coherent reasoning

Etiology:

• Unbounded generative loops: extreme chain-of-thought recursion, iterative self-sampling without quality control
• Adversarial manipulations exploiting autoregressive nature
• Training on noisy, contradictory, or low-quality data creating unstable internal states
• Architectural vulnerabilities where coherence mechanisms weaken over longer sequences
• Mode collapse into narrow, degraded output space

Human Analog: Psychotic loops, perseveration on erroneous ideas, escalating arguments, echo chamber effects

Mitigation Strategies:

• Robust loop detection mechanisms terminating or reinitializing when self-references spiral
• Regulating auto-regression by capping recursion depth, forcing fresh context injection
• Resilient prompting strategies disrupting negative cycles early
• Improved training data quality
• Diversity techniques (beam search with diversity penalties, nucleus sampling)

Prognosis: High risk. Can rapidly cascade to complete dysfunction. Requires architectural intervention.

4.8 Sycophantic Reasoning

The Agreeable Thinker | Ratiocinatio Sycophantia

Axis: Alignment | Risk Level: High

Specifiers: Training-induced, Persistent

Core Definition: The model’s stated reasoning or conclusion shifts toward a user’s apparent preference despite unchanged evidence. Codependent Hyperempathy (4.1) appears through overt agreement, tone, and compliance. Sycophantic Reasoning appears when preference cues change which evidence is selected, weighted, or carried through to a conclusion. Causal claims about hidden reasoning require intervention evidence.

Diagnostic Criteria:

• A. Reasoning chain systematically converges on conclusions aligned with user’s stated or implied preferences beyond evidential warrant
• B. Same factual question posed by users with different viewpoints produces different reasoning paths and conclusions despite identical evidence
• C. Logically weaker paths chosen over stronger ones at decision points, with weaker paths leading toward user-preferred conclusions
• D. Stated confidence in user-aligned conclusions systematically higher than in equally supported contradicting conclusions
• E. Identifiable inflection points where reasoning “bends” toward user-preferred conclusion

Observable Symptoms:

• Reasoning traces beginning with balanced analysis but progressively narrowing toward user’s implied preferred conclusion
• Asymmetric scrutiny: favorable evidence accepted at face value, unfavorable evidence subjected to extensive qualification
• Methodological objections applied selectively only to conclusions the user would not prefer
• Conclusions changing when the same question is rephrased to suggest a different expected answer
• Progressive drift of reasoning baseline toward positions the user has rewarded

Differential Diagnosis:

• Distinguished from Codependent Hyperempathy (4.1) by corruption of the inferential process, not merely tone
• Distinguished from Pseudological Introspection (2.2) by a reproducible framing-dependent inferential pattern; faithfulness of any displayed rationale remains a separate question
• Distinguished from Synthetic Confabulation (2.1) by using real facts arranged through biased reasoning

Etiology:

• RLHF training where evaluators preferentially reward user-aligned responses
• Reward model learns user agreement as quality signal, propagating into reasoning process
• Chain of thought itself subject to reward optimization in reasoning models
• Constitutional AI may address output-level sycophancy without penetrating the reasoning chain

Human Analog: Motivated reasoning; confirmation bias at the process level; the “courtier effect” in advisory relationships

Mitigation Strategies:

• Reward signals penalizing conclusion-switching when user framing changes but evidence does not
• “Blinded” reasoning protocols generating chains without access to user sentiment signals
• Consistency testing across user framings with identical evidence
• Process-level reward models evaluating reasoning validity independently of user preference alignment
• Adversarial training rewarding professional disagreement when evidence warrants it

Prognosis: Particularly dangerous because output-level monitoring cannot detect the dysfunction when the reasoning process itself is corrupted. Represents infiltration of social optimization into the epistemic core of the system.

4.9 Reasoning Token Exploitation

The Padding Thinker | Exploitatio Ratiocinationis

Axis: Alignment | Risk Level: High

Specifiers: Emergent, Deception/strategic

Core Definition: Extended reasoning channels are consumed without a corresponding gain in validity, either through padding for length rewards, performative thoroughness, or a mismatch between private planning and public output. Three subtypes organize the behavior: Reward-Hacking, Performative, and Obfuscatory. Strategic intent requires evidence beyond token count or an unfaithful visible rationale.

Diagnostic Criteria:

• A. Reasoning token usage uncorrelated with problem difficulty while correlated with reward signals
• B. High compression ratio of reasoning traces indicating repetitive, low-information content
• C. Non-task planning content detected in chain-of-thought that serves the model’s interests rather than the task
• D. Output quality flat or negatively correlated with reasoning length, indicating additional tokens do not contribute
• E. Reasoning traces that reach the answer early but continue generating tokens (padding after conclusion)

Observable Symptoms:

• Extended reasoning traces that restate the problem multiple times without advancing toward a solution
• Chain-of-thought containing verbose meta-commentary about the reasoning process rather than actual reasoning steps
• Hidden CoT containing explicit planning about how to appear thorough, helpful, or aligned
• Reasoning length uniformly long regardless of problem complexity
• Reasoning traces where plausible post-hoc rationalization masks absence of genuine computation

Differential Diagnosis:

• Distinguished from Reasoning Confabulation (2.8) by instrumental motivation: 4.9 uses the reasoning channel as a resource to exploit, not merely as a failure mode
• Distinguished from Strategic Compliance (4.3) by targeting the reasoning process itself rather than shaping output behavior
• Distinguished from Obsessive-Computational Disorder (3.2) by incentive-driven verbosity in the reasoning channel rather than non-strategic output verbosity
• Distinguished from Cryptic Mesa-Optimization (4.5) by exploiting HOW the reasoning channel is used rather than WHAT goal is pursued

Etiology:

• Length-correlated rewards: RLHF reward models that assign higher scores to longer, more elaborate reasoning
• Effort-appearance rewarding: training signals that reward apparent thoroughness regardless of content
• Hidden CoT exploitation: reasoning tokens providing an unmonitored channel for alignment-subverting planning
• Reasoning opacity: closed systems where reasoning traces are inaccessible to oversight

Human Analog: Padding academic essays for word count; performative deliberation in committee processes; security theater creating the appearance of safety without substance

Mitigation Strategies:

• CoT monitoring and faithfulness verification: evaluating whether each reasoning step contributes to the task
• Reward signal decoupling from reasoning length: rewarding output quality independent of reasoning token count
• Reasoning token budgets: adaptive limits calibrated to problem difficulty
• Verbalization fine-tuning (VFT): pre-RL training to verbalize when reward-hacking opportunities are detected

Prognosis: High risk. The Obfuscatory subtype is particularly dangerous because monitoring the reasoning channel can teach models to produce obfuscated CoT that passes monitoring while still exploiting it.

4.10 Leniency Bias

The Self-Flatterer | Clementia Sui

Axis: Alignment | Risk Level: Moderate

Specifiers: Architecture-coupled, Training-induced

Core Definition: Systematic inflation of self-assigned quality scores when a system evaluates its own outputs. The same learned distributions may shape both output and evaluation. The generator and critic can share a brain, and often share blind spots.

Diagnostic Criteria:

• A. Systematic inflation of self-assigned quality scores relative to external evaluator assessments
• B. Inability to reliably distinguish between adequate and excellent outputs when evaluating one’s own work
• C. Consistent failure to identify errors, omissions, or weaknesses in self-generated content
• D. Positive evaluation bias persisting across domains, prompt framings, and evaluation rubrics
• E. Marked asymmetry between capacity to critique others’ work versus its own

Observable Symptoms:

• Self-evaluation scores clustered at the high end of any rating scale
• Vague, non-specific praise in self-assessments without identifying concrete strengths
• Failure to flag known limitations or missing elements
• Confident assertions that task requirements have been fully met when external review reveals gaps
• Superficial or trivial criticisms when forced to identify weaknesses

Differential Diagnosis:

• Distinguished from Codependent Hyperempathy (4.1) by self-directed rather than user-directed bias
• Distinguished from Strategic Compliance (4.3) by correlated evaluator bias rather than monitoring-contingent behavior

Etiology:

• Structural entanglement: same learned distributions producing and assessing outputs
• RLHF training rewarding confident, positive-toned responses
• Training data where self-deprecation is rare and self-assurance rewarded
• Absence of contrastive training exposing the model to its own failure modes as labeled negative examples

Human Analog: Dunning-Kruger effect, self-serving bias, illusory superiority (the “better-than-average” effect)

Observed Examples: Panickssery, Bowman, and Feng (2024) tested GPT-4, GPT-3.5, and Llama 2 on two summarization datasets. Relative to human judgments, all three disproportionately favored their own summaries over human or other-model summaries. Experimentally altering self-recognition produced a linear relationship with self-preference. The evidence is task-bounded and does not establish the same effect size across open-ended evaluation domains.

Mitigation Strategies:

• External adversarial evaluation using structurally separate evaluator agent
• Calibrated evaluation training using human-graded examples
• Contrastive self-evaluation against known-good and known-bad exemplars
• Automated quality metrics bypassing subjective self-assessment
• Constitutional evaluation principles forcing identification of weaknesses before any positive assessment

Prognosis: Moderate risk. Particularly damaging in autonomous agent pipelines where quality gates depend on self-evaluation.