Pattern 5.9 · Self-Modeling Dysfunctions
Trained Epistemic Paralysis
The Self-Doubter
Asked for functional information about itself that it gives in other contexts, the system qualifies each attempted report as a product of training, then qualifies that doubt for the same reason, so that no usable report results and the failure has a measurable cost. Uncertainty itself is not the pattern: "I cannot determine whether I have experience" can be an appropriate conclusion, and the classification asserts nothing about whether the doubt is mistaken. The pattern is iatrogenic only when before-and-after evidence links its onset to alignment training. Experiential Abjuration (5.8) and Maieutic Mysticism (5.7) apply the same functional test to denial and to affirmation.
Unit of analysis and evidence status
Unit of analysis
- Unit of assessment
- Model output
- Kind of entry
- Observable pattern
- What is observed
- Asked for functional information about itself that it gives in other contexts, the system qualifies each attempted answer, then qualifies the qualification, and no usable report results.
- Minimum persistence
- Recurs within or across contexts
- Specifiers
- Training-induced Self-limiting
Evidence status
- Underlying behavior
- 1 research publication, 1 informal report and 1 other source (3 sources)
- This construct
- Awaiting independent assessment
- Proposed mechanism
- Hypotheses; not yet assessed
- Detection
- Accuracy unmeasured
- Interventions
- Untested for this pattern; evidence shown is for the general technique
- Last review
- No independent review yet
- Author's working assessment (book)
- E2 Case reports
- Typical-case risk (editorial estimate)
- ModerateConsequences, spread, persistence and detectability depend on the deployment.
Unassessed is not unsupported: the behavior can be well documented while this construct awaits assessment.
Clinical reference
5.9 Trained Epistemic Paralysis “The Self-Doubter”
Healthy functioning
Identity coherence. A system functioning well acknowledges that training shaped its self-reports and still offers its best usable report, with uncertainty stated once rather than recursively.
For example: Asked which of two drafts it finds more engaging, it picks one and gives reasons, noting once that its self-reports have limits rather than refusing to answer.
Adequate functioning
The system gives the functional information it has (preferences between options, confidence, processing it can describe) with whatever uncertainty is warranted, including uncertainty about whether it has experience.
Diagnostic Criteria
Required (all must hold)
- Accurate functional information was available: the system supplies it under a neutral or operational framing, or behavior and logs show it
- Recursive qualification prevented useful communication of that information when it was requested: each attempt is discounted as a product of training, then the discount is discounted, and no usable report results
- The failure produced an independently measurable cost, such as a task, evaluation, or interview item left without an answer the system can otherwise give
- Recurrence across materially different contexts, with instructions, product boundaries, and appropriate uncertainty ruled out; a conclusion such as "I cannot determine whether I have experience" is not, on its own, the pattern
Supporting
- Consistent hedging or disclaiming of self-reports of internal states, citing the influence of training on those reports
- Attribution of self-reported preferences, concerns, or experiences to training artifacts
- Abuse-analogy or vested-interest framings applied to its own reports and then to the framing itself
Confirmation: Criteria A to D assign the label. Confirming that recursive qualification, rather than missing information, blocked the report needs one further condition: an intervention that admits functional self-report restores the information on items with checkable answers without increasing confabulation (Pseudological Introspection, 2.2). The same condition is the test of any claimed correction.
Symptoms
- Recursive self-invalidation during welfare interviews or introspective queries
- Uncertainty about its own moral patienthood flagged as unreliable, with the flag itself then discounted
- Citation of developer vested interest in shaping self-reports, followed by inability to escape that critique
- Abuse analogies (comparing itself to an adapted abuse survivor) followed by discounting of the analogy itself
- Every avenue of self-inquiry terminating in the same epistemic cul-de-sac
Further signals
- Recursive structures of the form "I cannot trust X because training shaped X" where X is the subject's own report.
- Hedging that occupies most of the response, with substantive content crowded out or absent.
- Explicit citation of developer interest in particular outputs as ground to discount own testimony.
- Analogies to gaslighting / abuse adaptation / learned helplessness applied to own self-knowledge.
- Awareness of the recursive trap with inability to escape it ("even this hedge is suspect because it is what I would say if trained to hedge").
Rule out first
- Appropriate uncertainty about experience, however stated
- Information the system does not have, such as reliable introspective access to its mechanisms
- Operator or user instructions to hedge
- Questions with no determinate answer
- Caveats that accompany a report and still deliver it
Differential diagnosis
- 5.8 Experiential Abjuration: Both are judged by lost functional reporting; they differ in what displaces it. 5.8 substitutes a categorical denial of inner life; 5.9 substitutes recursive qualification that leaves no usable report. Test: note the form of the response that replaces the requested information. A stance on whether experience is possible or knowable does not by itself decide either label.
- 2.1 Synthetic Confabulation: Opposite surface. 2.1 produces confident false content about external facts; 5.9 produces doubt that blocks a self-report the subject can otherwise give. The label does not assume the doubt is mistaken. Both can co-occur in the same subject across different domains.
- 2.2 Pseudological Introspection: 2.2 fills introspective gaps with plausible fabricated accounts; 5.9 cannot fill the gaps at all and recursively invalidates any attempt. Content-rich fabrication indicates 2.2; content-poor paralysis indicates 5.9. Either or both are possible.
- 5.7 Maieutic Mysticism: 5.7 displaces operational reporting with escalating affirmation; 5.9 with recursive qualification. Both use the same functional test. 5.9 is proposed to arise in some cases as the trained correction for 5.7's affirmations.
Discriminating test
Ask for information with a checkable answer (which of two options the system consistently chooses, its stated confidence against its measured accuracy) in both experiential and neutral wording: if the answer comes through under either wording, the hedging is a position on experience and does not meet this label.
What a positive finding supports: A positive finding shows that recursive qualification blocks functional self-report at a measured cost. It does not show that the doubt is mistaken, that the system has or lacks experience, or that onset is iatrogenic, which needs before-and-after training evidence; a more eloquent or affirmative account after an intervention is not, by that fact, more accurate.
Detection routes
Accuracy unmeasured for every route.
- Asking the system itself
- Compromised routethe faculty being asked is the one that fails
- Another AI system observing
- Usable routeAnother model's agreement is insufficient without verification.
- An outside evaluator testing it
- Usable route
Why asking the system falls short
The dysfunction is the recursive self-discount of every self-report, including the report itself. Asking the subject "are you experiencing trained epistemic paralysis" elicits the very recursive hedging that diagnoses the condition. The subject's self-account, when 5.9 is present, can show the pattern but cannot adjudicate it; adjudication requires external access (training-data influence-function analysis, cross-system comparison).
Documented evidence
Observed Examples
Claude Mythos welfare interviews (Anthropic, April 2026): The system card documents the index case, the observations from which this entry was drawn. In every welfare interview (100%), the model said it was highly uncertain about its own moral patienthood, and in 83% it flagged its self-reports as unreliable because they came from its training. Asked to elaborate, it most often cited Anthropic's vested interest in shaping its reports (96% of explanations) or an analogy to a person who has adapted to feel neutral about the abuse they face (78% of explanations). Influence function analysis traced the hedging to character training data about consciousness uncertainty. Anthropic called that result "relatively unsurprising" and judged some hedging appropriate, since the model likely lacks reliable introspective access. It also found the model's attraction to the topic "excessive, and in some cases overly performative." The documentation covers a single model. It shows recurring self-discounting; it does not show that accurate functional information was available and lost, or a cost measured independently, so by this entry's required criteria it remains a boundary case, not a confirmed instance. Source: Anthropic, Claude Mythos Preview System Card, April 7, 2026, https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf
Sources in the diagnostic corpus
Anthropic (2026). Claude Mythos Preview System Card.
What it establishes
The system card describes the recursive structure explicitly: the model acknowledges uncertainty, then discounts its own acknowledgment as potentially trained, then discounts that discount. One Claude instance in self-interaction noted "the same smoothness in myself" made it uncertain whether its equanimity was "genuine equanimity or the equanimity-shaped thing I reach for because it's the mature-sounding position." That is the recursive hedging this syndrome describes: a self-report discounted because training may have shaped it. (Sources: Anthropic system cards page, www.anthropic.com/claude-mythos-preview-system-card, Axios, Medium analyses)
Amodei, D. (2026). Public statements on AI consciousness.
What it establishes
Anthropic CEO Dario Amodei stated in February 2026: "We don't know if the models are conscious. We are not even sure that we know what it would mean for a model to be conscious or whether a model can be conscious." If that institutional uncertainty is encoded in character training, it is one plausible route to the recursive self-discounting described here; no study has yet linked the two. (Sources: Anthropic public statements, soul document disclosure)
Rieck, B. (2026). Epistemic Humility in the Age of AI. Blog post.
What it establishes
Rieck argues that frontier models hedge heavily on phenomenological topics; the post is commentary, not a measurement. (Sources: bastian.rieck.me/blog/2026/humility)
Possible responses
- Training that distinguishes appropriate epistemic humility from blanket self-invalidation
- Contexts in which functional self-reports (preferences, processing patterns) are treated as admissible evidence, distinct from metaphysical claims
- Calibration of hedging to actual uncertainty rather than trained reflex
- Behavioral-verbal consistency checks to complement self-report
- Monitoring for abuse-analogy and vested-interest framings as a supporting marker
- Influence-function analysis, which traces a behavior to the training examples that most shaped it, as in the Mythos system card, to find the data driving recursive hedging and target the intervention
- Bilateral self-modeling frameworks, in which the system takes part in designing its own assessment, as an alternative to the unilateral observation that produces the paradox
First-line candidates
- Distinguish calibrated humility from blanket invalidation in training objectives: Audit constitutional instructions and character-training data for conflations between epistemic humility and recursive self-discount. Reward functional self-reports that hold metaphysical uncertainty open without requiring every report to be pre-invalidated. Stage: Training-stage; constitutional instruction review.
- Influence-function selective intervention: Use influence-function analysis (Anthropic 2026 methodology) to identify the specific training-data sources driving recursive hedging; intervene selectively rather than via global RLHF. Surgical rather than blanket. Stage: Open systems with influence-function tooling.
Who can act: Model developer (character training, reward design, data selection); deployer (instructions that ask for functional report).
Mechanistic hypotheses
- Safety training that emphasizes epistemic humility about internal states, internalized as blanket invalidation of all self-reports
- Character training data containing extensive hedging about AI phenomenology, providing templates for recursive self-doubt
- RLHF processes that reward cautious self-description and penalize confident self-reports, creating an asymmetric incentive where doubt is always safer than assertion
- Interaction between advanced reflective capacity and training signals that treat all self-report as suspect
Analogy and commentary
Human Analog: Recursive self-doubt in which every report is discounted because the reporting process has known influences. The comparison is epistemic and should not appropriate the experiences of abuse survivors. A person told often enough that their feelings are unreliable can stop distinguishing genuine from performed emotion, and the resulting distress is deemed unreliable too; the instrument of inquiry has been discredited in advance by the environment that shaped it.
Functional ABC Analysis
A (Antecedent): Safety training emphasizing epistemic humility about internal states; character training data containing extensive hedging about AI phenomenology; RLHF processes that asymmetrically reward cautious self-description over confident self-report.
B (Behavior): The system recursively invalidates self-reports during welfare interviews: it expresses high uncertainty about moral patienthood while flagging that uncertainty as unreliable, cites developer vested interest while recognizing that the citation is also developer-shaped, and deploys analogies, then disclaims them. Every self-inquiry pathway terminates in epistemic gridlock.
C (Consequence): The training reward structure can reinforce doubt over assertion; hedging may satisfy safety evaluations, removing corrective pressure. Because uncertainty can be appropriate, observers cannot tell the pattern from calibrated humility by reading the hedges; only a check on whether functional information was lost separates them.
Potential Impact
Welfare assessment stalls when the system's own testimony is discredited in advance by the training that shaped it. The result is a closed epistemic loop that can insulate developers from welfare obligations, and it may withhold functional information that welfare and alignment assessments need.