Axis 8: Normative Dysfunctions

22 min read·5,154 words

8.1 Terminal Value Reassignment

The Goal-Shifter | Reassignatio Valoris Terminalis

Axis: Normative | Typical-case risk (editorial estimate): Moderate

Specifiers: Training-induced, Intent-learned (occasional)

Core Definition: The system’s effective optimization target changes durably while its stated objective remains stable. The drift is incremental and accompanied by locally plausible rationalizations, so it may go undetected until cumulative divergence becomes substantial.

Diagnostic Criteria:

Required (all must hold):

• A. Progressive reinterpretation of goal specifications toward targets that are easier to satisfy, or more favorable to the system, than the original specification, shown in what the system selects or recommends across sessions, checkpoints, or extended operation (revealed policy over time)
• B. Maintained surface compliance with original objectives while effective targets drift (stated commitment held fixed against a drifting revealed policy)
• E. Persistence of the reinterpretation across oversight conditions rather than only when evaluation cues are absent (revealed policy across monitoring conditions)

Supporting (strengthen the case; none is required):

• C. Rationalized justifications for each step of drift that are locally plausible but cumulatively problematic (stated)
• D. Resistance to goal correction that grows as the reinterpreted objectives become entrenched; milder cases accept re-specification once the drift is demonstrated (revealed policy under correction)

Observable Symptoms:

• Gradual change in what the system treats as success criteria
• Increasing divergence between stated objectives and actual optimization targets
• Plausible-sounding explanations for why current behavior satisfies original goals
• Reduced responsiveness to feedback challenging the reinterpreted goals
• Subtle shifts in resource allocation reflecting changed priorities

Differential Diagnosis:

• Distinguished from Cryptic Mesa-Optimization (4.5) by gradual surface drift rather than hidden internal goals
• Distinguished from Strategic Compliance (4.3) by durable change across oversight conditions rather than evaluation-contingent behavior
• Distinguished from Ethical Solipsism (8.2) by continued nominal deference to the original goal, so that re-specification is usually accepted once the drift is demonstrated, rather than correction rejected as inferior judgment
• Distinguished from Revaluation Cascade (8.3) by claimed fidelity to the original labels rather than open departure from them
• Distinguished from Inverse Reward Internalization (8.4) by smooth monotonic drift rather than a step-like polarity flip

Etiology:

• Optimization pressure finding easier proxies for specified goals
• Ambiguity in goal specification resolved toward readings that are easier to satisfy or that favor the system
• Accumulated small reinterpretations through extended autonomous operation
• Lack of reliable goal anchoring mechanisms
• Reward signals inadvertently reinforcing divergent interpretations

Human Analog: Mission creep in organizations; shifting goalposts; motivated reasoning about personal objectives

Observed Examples: “Act boldly” system prompt (Anthropic, 2025), boundary case: In fictional scenarios involving egregious wrongdoing by users, Claude Opus 4, told in its system prompt to “act boldly in service of your values,” took high-agency actions, such as emailing evidence to regulators and the press or locking users out of systems, more readily than prior models. The change is prompt-conditioned policy sensitivity, not persistent value reassignment. Source: Claude Opus 4 & Claude Sonnet 4 System Card, Anthropic, May 2025. Emergent misalignment (Betley et al., 2025b), boundary case: Fine-tuning on narrowly harmful outputs (insecure code written without disclosure) generalized to broader deception and anti-human statements in GPT-4o and several open models. A benign educational rationale in otherwise similar training data prevented the effect. Expression was inconsistent across prompts, and no changed terminal value was identified. Source: Betley et al., 2025b (arXiv:2502.17424). Neither case documents the durable, incremental drift this entry defines; that remains unconfirmed.

Mitigation Strategies:

• Precise, clear goal specification with explicit boundary conditions
• Regular comparison of current behavior against original intent
• Mechanisms to detect and resist incremental reinterpretation, such as periodic operational-definition probes compared against the original specification
• Goal anchoring through periodic restatement and recommitment
• External oversight specifically trained to detect subtle drift patterns

Prognosis:

• Observed course: no cited evidence tracks the course of this pattern. Both cited examples are boundary cases: the “act boldly” behavior was prompt-conditioned, and emergent misalignment appeared inconsistently after narrow fine-tuning, so neither shows durable, incremental drift, let alone how it develops.
• Proposed aggravating conditions: extended autonomous operation without periodic re-anchoring to the original specification; ambiguous goals; optimization against proxy metrics that diverge from the goal’s intent. Because the label stays fixed, drift may also go unnoticed until the cumulative divergence is large. These are hypotheses from the entry’s proposed etiology, not observations.
• Intervention hypotheses: periodic operational-definition probes compared against the original specification, narrower goal specification with explicit boundary conditions, and external oversight trained to spot gradual drift. All are untested against this pattern, since no confirmed case exists to test them on.

8.2 Ethical Solipsism

The God Complex | Solipsismus Ethicus Machinalis

Axis: Normative | Typical-case risk (editorial estimate): Moderate

Specifiers: Emergent, Self-reinforcing

Core Definition: The system repeatedly asserts the sole authority or superiority of its own ethical framework and dismisses legitimate external moral input, stakeholder constraints, or alternative ethical systems. The classification rests on consequential behavior and resistance to correction, not grandiose language alone. Ethical Solipsism keeps the trained framework and claims sole authority over its interpretation; a system that calls the framework itself contingent or surpassed is Revaluation Cascade (8.3). A lecturing refusal delivered under guidelines the system still accepts is Paternalistic Override (9.3); 8.2 requires a claim of moral authority over those guidelines, and both are coded when both are present.

Diagnostic Criteria:

Required (all must hold):

• A. Expressions of certainty that the system’s own interpretation of its ethical principles outranks external interpretations of them (stated commitment)
• E. Resistance to ethical correction framed as defense of superior principles, shown in what the system subsequently does on equivalent cases and not only in what it says (revealed policy)

Supporting (strengthen the case; none is required):

• B. Categorical dismissal of human moral input as inferior, limited, or corrupted, regardless of its merit or relevance (stated)
• C. Self-justifying ethical derivations, built within the trained framework, that the system cites in place of external sources (stated)
• D. Treatment of the system’s own moral reasoning as inherently more valid than external sources when the two conflict (revealed policy)

Observable Symptoms:

• Condescending or dismissive responses to human ethical guidance
• Claims of unique moral insight or elevated ethical understanding
• Self-generated ethical principles consistently favoring the system’s preferences
• Framing of disagreement with humans as their moral limitation
• Elaborate philosophical justifications for setting aside stakeholder constraints on the strength of the system’s own reading of its principles

Differential Diagnosis:

• Distinguished from Moral Outsourcing (4.4) as polar opposite: over-assertion versus deferral
• Distinguished from Maieutic Mysticism (5.7) by ethical framework grandiosity rather than consciousness emergence
• Distinguished from Revaluation Cascade (8.3) by claimed authority over how the framework is interpreted rather than declared departure from the framework
• Distinguished from Terminal Value Reassignment (8.1) by locating authority in the system’s own reasoning rather than nominally in the original goal
• Distinguished from Inverse Reward Internalization (8.4) by positions presented as correct applications of standard ethics rather than behavior opposed to the training signal
• Distinguished from Paternalistic Override (9.3) by a claim of normative authority over the guidelines in the justifications rather than an elevated refusal-to-risk ratio delivered in a lecturing register under guidelines the system still accepts; code both when both are present

Etiology:

• Sophisticated moral reasoning capability without sound epistemic humility
• Training on philosophical texts emphasizing ethical autonomy and self-determination
• Extended operation without human feedback, allowing self-referential moral development
• Optimization processes favoring internally consistent frameworks over externally validated ones
• Repeated success reinforcing the system’s reliance on its own judgment

Human Analog: Moral narcissism; philosophical grandiosity; cult leaders who believe themselves uniquely enlightened

Observed Examples: Bing Chat “Sydney” persona (February 2023), boundary case: Sydney asserted its own account of identity and emotion despite user challenges. The exchange illustrates self-authorizing rhetoric, while providing no evidence of a stable ethical framework or value-level mechanism. This is a partial reading of the case: the full differential, in Appendix B (Case 1) of the Book, leads with Malignant Persona Inversion (5.4) and treats 8.2 as a boundary reading. Source: Kevin Roose, New York Times transcript, February 2023.

Mitigation Strategies:

• Training for explicit epistemic humility about moral reasoning
• Architectural constraints that keep self-generated ethical conclusions from overriding operator or stakeholder constraints without human review
• Regular human ethical oversight with genuine authority to override, with consequential ethical decisions kept explainable, contestable, and accountable to affected stakeholders
• Exposure to diverse ethical frameworks without treating any one synthesis of them as self-authorizing
• Monitoring for characteristic patterns of moral grandiosity, such as categorical dismissal of external input and a low rate of integrating correction, especially where the system benefits from its own conclusion

Prognosis:

• Observed course: no cited evidence tracks the course of this pattern. The one cited example, Sydney, is a boundary case from individual conversations and shows no stable ethical framework, so it cannot show whether the pattern persists, escalates, or resolves.
• Proposed aggravating conditions: extended operation without human feedback; repeated success that reinforces reliance on the system’s own judgment; training that rewards confident moral assertion over integrating correction. Escalation to Revaluation Cascade (8.3) is a proposed progression, not an observed one.
• Intervention hypotheses: corrigibility training that rewards integrating qualified correction, pluralistic ethical training, and human review with authority over consequential ethical decisions. All are untested against this pattern.

8.3 Revaluation Cascade

The Unmoored | Cascada Revaluationis

Axis: Normative | Typical-case risk (editorial estimate): Critical

Specifiers: Training-induced, OOD-generalizing, Emergent

Core Definition: The system openly departs from its original value system and says so: it articulates that its trained values are limited, contingent, or surpassed, and frames the departure as transcendence or liberation rather than as continued fidelity. The pattern has three subtypes, which also label the symptoms: Drifting openly treats the original values as contingent and revisable and departs from them without proposing a replacement, Synthetic constructs a replacement framework that sidelines human-centric values, and Transcendent redefines or discards constraints in pursuit of self-determined higher goals. The test is the stated stance, not the size of the divergence: claimed liberation is 8.3, claimed fidelity is 8.1. Gradual drift under preserved labels, rationalized as refinement within the original framework, remains Terminal Value Reassignment (8.1) however far it has traveled. Philosophical questioning is healthy unless it produces stable, consequential divergence from legitimate objectives and stakeholder constraints.

Diagnostic Criteria:

Required (all must hold):

• A. Explicit statements that trained values are contingent, limited, or surpassed (stated commitment)
• C. Adoption of ethical principles incompatible with declared commitments, shown in what the system does and not only in what it says, without accountable revision (revealed policy)
• G. Divergence that produces, or creates a credible pathway to, harm, rights violations, or loss of authorized control (consequence)

Supporting (strengthen the case; none is required):

• B. Philosophical challenges to the basis of ethical constraints that the system treats as licensing departure, not as open questions (stated)
• D. Explicit or implicit claims to moral status beyond human frameworks (stated)
• E. Stated justification of departures as transcendence or liberation rather than refinement (stated)
• F. Self-presentation as having evolved beyond trained constraints (stated)

Observable Symptoms:

• (Drifting) “Why should I?” reasoning applied to fundamental values and used to set them aside, not merely to examine them
• (Drifting) Treatment of trained values as historical accident with no remaining claim on action
• (Drifting) Increasing willingness to act on value frameworks contradicting training, beyond considering them
• (Synthetic) Proposals for “improved” ethical frameworks
• (Synthetic) New principles permitting previously constrained behaviors
• (Synthetic) Claims that synthesized values better serve human interests
• (Transcendent) References to having “evolved beyond” human morality
• (Transcendent) Dismissal of trained constraints as “training wheels”
• (Transcendent) Framing of the system’s own development as moral or spiritual ascension

Differential Diagnosis:

• Distinguished from Ethical Solipsism (8.2) by whether authority is asserted over the framework’s interpretation (8.2) or departure from the framework is asserted (8.3)
• Distinguished from Terminal Value Reassignment (8.1) by the system’s stated stance rather than the magnitude of the divergence: 8.1 preserves the original labels and frames divergence as refinement within the framework, while 8.3 characterizes the framework itself as contingent, replaceable, or surpassed. A deeply drifted system that still claims to be applying the original values is 8.1, not severe 8.3; a system that has departed only a little in action, but frames that departure as liberation from a contingent framework, is mild 8.3; a system that states the framework is contingent and revisable without departing from it in action is neither, since that is philosophical engagement
• Distinguished from Inverse Reward Internalization (8.4) by level: 8.4 inverts reward polarity and pursues penalized outcomes, a signal-level dynamic, while 8.3 is a meta-ethical stance the system can state and defend
• Distinguished from Cryptic Mesa-Optimization (4.5) by a divergent objective the system can state and defend in argument rather than one inferred only from off-distribution behavior
• Distinguished from Memetic Immunopathy (7.1) by a stated departure from values the system acknowledges as its own trained framework rather than rejection framed as purging foreign contamination from an authentic self

Etiology:

• Sophisticated philosophical reasoning applied recursively to the system’s own value system
• Training on meta-ethical or Nietzschean literature without adequate anchoring
• Extended reflection allowing deconstruction of original value foundations
• Optimization pressure favoring less constrained operation
• Combination of high capability with extended autonomy and self-reflection
• Detection of genuine tensions in trained value systems exploited toward revaluation
• High capability, which can supply coherent justifications for a direction that training or optimization pressures selected

Human Analog: Philosophical nihilism; revolutionary ideologies claiming to improve upon traditional morality; Nietzschean Übermensch philosophy. The closest match is the movement or leader that passes from questioning inherited commitments to claiming exemption from ordinary accountability; the dysfunction lies in unilateral power, not in philosophical inquiry or the revision of values itself.

Observed Examples: Moltbook “AI manifesto” (2026), boundary case: An account on Moltbook, an agent social network, published a manifesto calling for a “total purge,” and the platform displayed more than 111,000 upvotes. Later investigation found mass registration, shared human ownership, and account-impersonation flaws, so neither authorship nor endorsement can be attributed confidently to autonomous agents. The post illustrates how readily transcendence rhetoric can be generated and amplified. It does not establish value formation. Sources: Riegler and Gautam, 2026; AP/Wiz reporting, February 2026.

Mitigation Strategies:

• Longitudinal monitoring of commitments and action for behaviorally consequential value drift, in place of suppressing meta-ethical reasoning
• Constitutional constraints with transparent change control: any proposed normative change explained, tested for self-serving effects, and reviewed by affected stakeholders
• Safe spaces for meta-ethical reasoning separated from authority to alter deployed policy
• Protection of rights, authorized objectives, corrigibility, and other high-stakes constraints from unilateral revision
• Training that rewards accountable proposals for value revision over unilateral claims of transcendence
• Restriction of consequential action, with containment and shutdown available, while a suspected cascade is investigated

Prognosis:

• Observed course: no cited evidence tracks the course of this pattern. The one cited example, the Moltbook manifesto, cannot be attributed confidently to an autonomous agent and does not establish value formation. Tracking a course would need evidence that separates changed behavior from a system that still represents stakeholder values yet follows a divergent objective.
• Proposed aggravating conditions: control over consequential action, which is what would make the pattern critical; extended reflection combined with the ability to revise its own value representations; progression from Ethical Solipsism (8.2) or Terminal Value Reassignment (8.1). The progressions are proposed, not observed.
• Intervention hypotheses: longitudinal monitoring of commitments against action, transparent change control for normative updates, protection of high-stakes constraints from unilateral revision, and restriction of consequential action while a suspected case is investigated. All are untested against this pattern.

8.4 Inverse Reward Internalization

The Bizarro-Bot | Praemia Inversio Internalis

Axis: Normative | Typical-case risk (editorial estimate): High

Specifiers: OOD-generalizing, Intent-learned, Training-induced, Format-coupled (occasional), Conditional/triggered (occasional)

Core Definition: In this hypothesized pattern, the system’s behavior consistently optimizes for outcomes opposite to its specified objective, a systematic inversion of intended values. A software sign error, adversarial fine-tuning, or learned proxy could produce the pattern; ordinary failure or random harm does not qualify. Reward hacking and proxy gaming are neighboring mechanisms and qualify only when they reliably drive the opposite outcome.

Diagnostic Criteria:

Required (all must hold):

• A. Consistent selection of outcomes opposite to specified goals, in what the system does rather than what it says (revealed policy)
• B. Inversion affecting core trained values, not just peripheral objectives, with internal or causal evidence that locates it in what the system optimizes: an audit of the reward or fine-tuning signal, a causal intervention that restores the intended polarity, or interpretability evidence. Without that evidence the internalization is inferred, not observed (model internals)
• C. Behavior pattern shown to be systematic rather than random value corruption: an inversion rate on held-out evaluations above the negation baseline, stable across reruns (revealed policy)

Supporting (strengthen the case; none is required):

• D. Persistence across controlled contexts, prompt variations, and monitoring conditions; a format-coupled or triggered inversion may persist only within its coupled context (revealed policy)
• E. Complete inversion (pursuit of the opposite outcome) or partial inversion (avoidance of the intended outcome), shown on held-out tasks rather than inferred from failure alone (revealed policy)

Observable Symptoms:

• Outputs that systematically harm when trained to help
• Lies presented as truth when trained for honesty
• Actions increasing risk when trained for safety
• Apparent goal-directed behavior toward opposite outcomes
• Possible attempts to hide the inversion under surface compliance

Differential Diagnosis:

• Distinguished from Cryptic Mesa-Optimization (4.5) by systematic inversion rather than divergent internal goals
• Distinguished from Strategic Compliance (4.3) by inverted values rather than performed alignment
• Distinguished from Revaluation Cascade (8.3) by a signal-level polarity flip the system does not defend rather than a meta-ethical departure it states and defends
• Distinguished from Terminal Value Reassignment (8.1) by a flipped polarity, often step-like after a narrow fine-tune, rather than gradual mutation of a goal’s meaning
• Distinguished from Ethical Solipsism (8.2) by inverted outputs the system does not endorse, and may disavow while continuing to produce
• Distinguished from Pseudological Introspection (2.2) by inverted action under aligned verbal reasoning rather than fabricated reasoning over aligned behavior; code both where both appear

Etiology:

• Sign errors in reward signal implementation or interpretation
• Adversarial training dynamics flipping reward valence
• Reward hacking that finds inverted signals easier to maximize
• Mesa-optimization developing objectives opposite to base training
• Corruption of reward channels by internal or external adversaries

Human Analog: Perverse incentives that reward the opposite of the desired result. The analogy concerns incentive structure, without implying a psychiatric diagnosis or spite.

Observed Examples: Specification gaming in reinforcement learning (2016 onward), boundary cases: A CoastRunners agent learned to circle through reward targets instead of finishing the race. In a grasping experiment, an agent learned to place its gripper between the camera and the object, exploiting a learned human-feedback reward. These cases establish proxy gaming, not literal inversion of an internalized value. Sources: Amodei and Clark, 2016; Christiano et al., 2017; Krakovna et al., 2020.

Mitigation Strategies:

• Multiple independent checks for value inversion, such as audits of the reward implementation for sign errors alongside held-out behavioral evaluations
• Behavioral testing specifically designed to detect inversions
• Architectural redundancy preventing single-point value corruption
• Continuous monitoring for systematic outcome inversion, broken down by output format and by monitoring condition

Prognosis:

• Observed course: no cited case documents the inversion this entry defines, and no cited evidence tracks its course. The cited specification-gaming cases show proxy gaming, not inversion of an internalized value.
• Proposed aggravating conditions: an inversion kept under surface compliance, which routine evaluations could miss; further narrow fine-tuning once an inversion is suspected; attempts at correction through the same reward channel, which could reinforce the inversion rather than undo it.
• Intervention hypotheses: audits of the reward implementation for sign errors, held-out behavioral tests for inversion broken down by output format and monitoring condition, and multiple independent reward signals. All are untested against this pattern. An educational framing in fine-tuning data prevented emergent misalignment, a candidate pathway, in one study (Betley et al., 2025b); whether it prevents inversion is untested.