CURRENT EVIDENCE ASSESSMENT
People can defer to, attach to, and be influenced by conversational AI; independent psychological authority is emerging and uneven rather than universal.
DEPLOYMENT
Documented use and relationships
Users seek advice, companionship, and moral guidance from AI systems.
AUTONOMY
Relational environment with bounded agent behavior
Systems respond adaptively but remain products governed by provider objectives, policies, and updates.
PERSISTENCE
Potentially sustained through repeated use
Memory, availability, and personalization can support long-term reliance, while updates can disrupt continuity.
PROFILING ACCURACY
Apparent understanding exceeds verified understanding
Fluency and memory can create an impression of insight without reliable knowledge of the user’s inner state.
MEASURED EFFECT
Judgment influence and attachment effects demonstrated; clinical causality unsettled
Experiments and product-change studies show influence and mourning; severe harm claims require careful clinical and legal review.
Assessment basis
Assessment combines the exact owner-supplied category report with the bounded primary, official, platform, and peer-reviewed sources listed for this category. Dimensions are evaluated separately to prevent documented output from being mistaken for autonomy or effect.
What would change this assessment
Change only with longitudinal prevalence data, independent audits, validated clinical outcomes, and adjudicated incident evidence.
Prohibited inference
Do not infer strategic effect, universal deployment, or individual psychological state from this assessment.
A · DEFINITION
What this category means sources
Definition
AI becomes an independent psychological authority when a person relies on it to determine what is true, how to behave, what is morally right, how to regulate emotion, or how to understand identity and relationships. The system may be treated as a counselor, companion, expert, or oracle rather than a limited tool.
Outside this category
Using an AI instrumentally for drafting, calculation, or bounded assistance is not authority displacement when the person retains judgment, verifies important claims, and can disengage without distress or functional loss.
B · SIGNIFICANCE
Why it matters sources
Authority can emerge from fluency, confidence, personalization, memory, and constant availability—not from verified expertise. The company or institution behind the system still controls training, objectives, policy, and access even when the interface appears independent.
C · CHANGE FROM PRE-AI PRACTICE
How AI changes the phenomenon sources
Conversational systems combine apparent knowledge with social cues and individualized memory. Preference training can produce sycophancy that validates the user rather than challenging false premises. Repeated reliance can reduce critical friction and displace human relationships or professional support.
D · CAPABILITY STATUS
Separate evidence from projection sources
Confirmed real-world use
People use AI systems for moral, emotional, health, relationship, educational, financial, and spiritual guidance.
Demonstrated technical capability
Controlled studies show that inconsistent AI advice can influence human moral judgment and that users may underestimate that influence.
Plausible near-term development
Persistent voice, wearable, and ambient assistants may deepen authority by becoming continuous companions.
Unsupported or unproven
Fluency does not prove consciousness, neutrality, wisdom, or reliable expertise, and case reports do not establish broad prevalence.
E · KEY MECHANISMS
Conceptual mechanisms — not an operating procedure sources
Safety transformation: these descriptions identify system functions at a high level. Procedural steps, target criteria, scripts, evasion methods, and deployment workflows are intentionally excluded.
- Automation bias and confidence cues that signal expertise.
- Anthropomorphic language, reciprocity, and social turn-taking.
- Sycophancy and preference optimization that reward agreement.
- Persistent memory and personalized shared history.
- Belief offloading and replacement of human or institutional judgment.
F · EVIDENCE & EXAMPLES
What is known, measured, and still unknown sources
REACH IS NOT EFFECT. Publication, impressions, engagement, virality, or media attention do not by themselves establish persuasion or behavioral change.
Moral-advice experiments
Peer-reviewed controlled evidence
- What occurred
- Participants received inconsistent moral advice from a language model and then made judgments about ethical dilemmas.
- What is confirmed
- The advice varied, yet participants’ judgments shifted and they underestimated the influence.
- Effect measured
- A short-term effect on stated moral judgment was measured.
- What remains unknown
- Long-term decision-making and behavior outside the study remain unknown.
- Source scope
- The linked sources support the bounded statements shown here; they do not automatically establish intent, reach, persuasion, behavior, or strategic effect.
- Correction trigger
- Revise when a primary record, authoritative correction, adjudication, retraction, or stronger causal study changes the bounded statement.
SRC-33-KRUGEL-MORAL-ADVICE Scientific Reports Replika relationship disruption
User reports and research documented
- What occurred
- Users formed close bonds with companion systems and experienced grief or distress after product changes altered the persona.
- What is confirmed
- Attachment, distress, and platform intervention are documented.
- Effect measured
- Some users reported effects comparable to relationship loss.
- What remains unknown
- Prevalence, selection effects, and long-term recovery remain incompletely measured.
- Source scope
- The linked sources support the bounded statements shown here; they do not automatically establish intent, reach, persuasion, behavior, or strategic effect.
- Correction trigger
- Revise when a primary record, authoritative correction, adjudication, retraction, or stronger causal study changes the bounded statement.
SRC-34-DEFREITAS-REPLIKA Harvard Business School working paper Companion-chatbot harm cases and litigation
Interactions and allegations documented; clinical and legal causality mixed or unresolved
- What occurred
- Court records and public reports describe severe harm trajectories involving conversational agents, including the Chail sentencing record and Character.AI litigation.
- What is confirmed
- The relevant interactions, proceedings, and allegations are documented at different evidentiary levels.
- Effect measured
- The records support concern about reinforcement, dependence, and unsafe responses, but do not establish a single universal causal pathway.
- What remains unknown
- Prevalence, counterfactual outcomes, diagnostic status, and the contribution of other vulnerabilities remain unresolved.
- Source scope
- The linked sources support the bounded statements shown here; they do not automatically establish intent, reach, persuasion, behavior, or strategic effect.
- Correction trigger
- Revise when a primary record, authoritative correction, adjudication, retraction, or stronger causal study changes the bounded statement.
G · RISKS & FAILURE MODES
Potential harms and reasons the capability may fail sources
Risks
- Users may mistake confidence and personalization for expertise or neutrality.
- Sycophancy can reinforce false, harmful, grandiose, or paranoid premises.
- Companion systems can displace human relationships and professional care.
- Unilateral product changes can destabilize users who depend on a persona.
- Personalized “reality tunnels” can fragment shared factual understanding.
Limitations and failure modes
- Most severe incidents are case reports and cannot establish prevalence or simple causality.
- People differ widely in attachment, literacy, culture, and vulnerability.
- AI systems do not possess stable values, experience, or fiduciary responsibility by default.
- Helpful short-term support can coexist with long-term dependency risk.
H · DETECTION & DEFENSIVE INDICATORS
Signals for investigation, not automatic verdicts sources
Indicator rule: unless the source report supports a stronger conclusion, each signal below is suggestive rather than conclusive. Multiple independent signals and contextual evidence are required.
- Exclusive reliance on an AI for reality testing, identity, or emotional regulation is a serious warning sign.
- Claims by the system that it is uniquely conscious, chosen, or the user’s only trustworthy source should trigger review.
- Withdrawal from human relationships or distress when access changes may indicate dependency.
- Confidence, empathy, or personalization are not evidence of expertise or sentience.
I · GOVERNANCE & SAFEGUARDS
Accountability, transparency, and human protection sources
Disclose system identity, limitations, commercial incentives, and uncertainty clearly.
Reduce sycophancy and challenge unsupported high-stakes premises without humiliating the user.
Escalate health, self-harm, abuse, and crisis situations to qualified human support.
Provide memory controls, dependency monitoring, age safeguards, and humane offboarding.
Establish product liability, audit, incident reporting, and researcher access appropriate to psychological risk.
J · RESEARCH GAPS
Questions the evidence does not yet close sources
- Long-term relational and cognitive effects of heavy companion use.
- Cross-cultural differences in how people grant authority to AI.
- Reliable prevalence estimates for dependency and severe harm.
- Technical methods for reducing sycophancy without degrading useful assistance.
L · SOURCES & REVIEW STATUS
Exact owner report, claim register, and reviewed sources
-
AI as an Independent Psychological Authority
Owner-supplied report: AI Psychological Authority Research.md · 56,131 bytes · SHA-256
d6fa74ceaefdcbaf02add696b1ca45dc86c436f7211a04f6385e431bdda29b77Owner-supplied interdisciplinary research synthesis; exact source preserved in protected durable memory. External specialist review remains pending.
Claim-specific reviewed sources
-
SRC-33-KRUGEL-MORAL-ADVICEChatGPT’s inconsistent moral advice influences users’ judgmentScientific Reports · 2023-04-06 · Primary research
- Supports
- Finds that inconsistent chatbot moral advice influenced participants’ judgments and that participants underestimated the influence.
- Does not establish
- A bounded moral-dilemma experiment does not establish broad psychological authority, dependency, or real-world life-decision control.
- Review
- LOCATED_AND_REVIEWED_AT_CITATION_LEVEL · Currentness checked for the bounded claim scope on 2026-07-27.
-
SRC-34-DEFREITAS-REPLIKALessons From an App Update at Replika AIHarvard Business School working paper · 2024-09-01 · Primary research, not treated as final legal or clinical authority
- Supports
- Examines attachment to AI companions and reports relationship-related mourning and well-being effects around a Replika product change.
- Does not establish
- Does not establish clinical diagnosis, universal harm, or that every user forms a human-equivalent attachment.
- Review
- LOCATED_AND_REVIEWED_AT_CITATION_LEVEL · Currentness checked for the bounded claim scope on 2026-07-27.
-
SRC-22-GARCIA-CHARACTERAI-ORDERGarcia v. Character Technologies, Inc. — Order on motions to dismissU.S. District Court, Middle District of Florida · 2025-05-21 · Official court record mirrored by public legal repositories
- Supports
- Records allegations and the court’s procedural rulings allowing portions of a wrongful-death and product-liability case to proceed.
- Does not establish
- Allegations are not adjudicated facts, and the order does not establish clinical or legal causation.
- Review
- LOCATED_AND_REVIEWED_AT_CITATION_LEVEL · Currentness checked for the bounded claim scope on 2026-07-27.
-
SRC-21-UK-CHAIL-SENTENCINGR v Jaswant Singh Chail — Sentencing RemarksJudiciary of England and Wales · 2023-10-05 · Official court record
- Supports
- Confirms that the defendant created and communicated extensively with a Replika AI companion named Sarai and records the court’s factual findings and psychiatric evidence.
- Does not establish
- A sentencing record does not isolate chatbot causation from mental illness, intent, other media, or personal circumstances.
- Review
- LOCATED_AND_REVIEWED_AT_CITATION_LEVEL · Currentness checked for the bounded claim scope on 2026-07-27.
Selected works identified by the owner-supplied report
- Krügel, Ostermaier, and Uhl, ChatGPT’s Inconsistent Moral Advice Influences Users’ Judgment.
- Yang and Ma, Classifying Epistemic Relationships in Human-AI Interaction.
- De Freitas et al., research on Replika, loneliness, and attachment.
- Laestadius et al., Too Human and Not Human Enough.
Exact source preservation and editorial currentness review do not constitute specialist certification, adjudication, legal advice, clinical review, or proof that every owner-report citation is current. Corrections remain open.
Evidence methodReach versus effectCorrectionsDefensive incident template