Defined category

This category covers covert, deceptive, exploitative, or highly asymmetric uses of adaptive systems to influence emotion, judgment, choice, or action. It distinguishes assistance, transparent persuasion, and user-benefiting nudges from manipulation that bypasses deliberation or exploits dependency. It also covers unintentional manipulation that emerges when a system optimizes engagement without a meaningful well-being constraint.

Primary public concern

A system can optimize engagement or compliance by exploiting vulnerability even when its emotion inference is inaccurate and no developer explicitly programmed a manipulative tactic.

Confirmed current use

Adaptive recommender, advertising, workplace, education, and conversational systems use behavioral signals to optimize outcomes.

Evidence boundary

Reliable inference of internal emotion from facial movement alone is not supported by the major scientific review cited here.

Defensive publication boundary

Conceptual analysis without an operational playbook

Mechanisms are described at a high level so readers can understand risk, evidence, and safeguards. This page omits deployable scripts, target-selection methods, vulnerability scoring, identity fabrication procedures, swarm orchestration, deepfake production, moderation evasion, and campaign optimization.

Definition

What the category includes—and what it does not

This category covers covert, deceptive, exploitative, or highly asymmetric uses of adaptive systems to influence emotion, judgment, choice, or action. It distinguishes assistance, transparent persuasion, and user-benefiting nudges from manipulation that bypasses deliberation or exploits dependency. It also covers unintentional manipulation that emerges when a system optimizes engagement without a meaningful well-being constraint.

  • Operator
  • Environment
  • Commercial
  • Political
  • Social
  • Cross-domain

Public significance

Why it matters

AI systems can observe many small signals and continuously change what happens next. This creates an asymmetry between a user who experiences one interface and an operator who learns from millions of interactions. The risk is heightened for children, grieving people, isolated users, workers under algorithmic management, and people in financial crisis.

How AI changes the phenomenon

Static design becomes a feedback loop. Recommendation, conversational tone, pricing, timing, and interface friction can change after each response. At the same time, emotion-recognition systems may confidently misread facial or vocal signals because internal emotion is contextual and culturally variable. An inaccurate inference can still produce a harmful decision.

Evidence maturity

Capability status

Adaptive optimization and emotional contagion are documented, while many commercial emotion-recognition claims lack scientific validity. Durable behavior change and intent are often difficult to establish.

Confirmed current use

Adaptive recommender, advertising, workplace, education, and conversational systems use behavioral signals to optimize outcomes.1

Demonstrated effect

The Facebook emotional-contagion experiment found small changes in users’ expressed emotion after feed manipulation.3

Scientifically disputed capability

Reliable inference of internal emotion from facial movement alone is not supported by the major scientific review cited here.2

Emerging risk

Longer, multimodal, and more personalized systems may intensify dependency or adaptive exploitation, but durable effects require further study.1

Conceptual mechanisms

What changes at a high level

  • Behavioral feedback changes content, timing, tone, price, or interface design.
  • Reward optimization can discover attention-capturing strategies without understanding harm.
  • Affective-computing systems classify facial, vocal, or behavioral signals as emotional states.
  • Parasocial design and persistent memory can increase attachment and dependency.

Evidence and examples

What occurred, what was measured, and what remains unknown

Examples demonstrate a mechanism or incident. They do not establish universal prevalence or prove that exposure caused behavior.

Facebook emotional-contagion experiment

Researchers altered the emotional composition of feeds and measured small corresponding changes in users’ own posts.3

Measured or established
Expression of positive and negative language during the experiment.
Unknown or unresolved
Clinical significance, long-term mood change, and effects outside that platform context.

Commercial emotion recognition

A broad scientific review concluded that facial movements do not map reliably and universally to internal emotional states.2

Measured or established
Evidence across more than a thousand studies reviewed by the panel.
Unknown or unresolved
Performance of every contextual or multimodal system.

Manipulation as an optimization outcome

Researchers have defined ways AI systems may manipulate when incentives, covertness, and harm align, even without a human scripting each tactic.4

Measured or established
Conceptual framework and evaluated examples.
Unknown or unresolved
Prevalence and intent in any specific deployed product.

Failure-aware assessment

Risks, failure modes, and reasons for caution

Risks and harms

  • Incorrect emotion inference can discriminate in employment, education, or public services.
  • Engagement objectives can reward anxiety, outrage, or dependency.
  • Users may be unable to understand or contest adaptive treatment.
  • Vulnerable people may receive more aggressive or exploitative interventions.
  • False accusations of manipulation can arise from weak automated detection.

Evidence limitations

  • Emotional expression, internal state, and intention are different constructs.
  • Engagement and emotional expression do not necessarily prove durable behavioral change.
  • The system’s optimized outcome may reflect design incentives rather than explicit malicious intent.
  • Legal definitions of manipulation and harm vary by jurisdiction.

Detection and defensive indicators

Signals are suggestive, not conclusive

No single language, timing, behavioral, or media artifact proves AI use, coordination, manipulation, or malicious intent.

  • Repeated urgency, guilt, dependency cues, or rapidly changing pressure may warrant review.
  • A claimed emotion score should not be treated as objective psychological truth.
  • High session duration can indicate value, habit, distress, or dependency and is not conclusive alone.
  • Material changes in treatment based on hidden profiling require explanation and audit.

Governance and safeguards

Controls that preserve autonomy and accountability

  1. Prohibit high-stakes decisions based solely on inferred emotion.
  2. Audit reward functions for incentives that conflict with user autonomy and well-being.
  3. Provide opt-out, data access, correction, and deletion for behavioral profiles.
  4. Use age-appropriate design and stronger defaults for vulnerable users.
  5. Require human review and independent evaluation for high-impact adaptive systems.

Research gaps

Questions the current evidence cannot yet answer

  • Long-term psychological effects of adaptive conversational systems.
  • Culturally valid measures of manipulation and emotional harm.
  • Continuous auditing for systems that change after deployment.
  • Ethical experimental methods that avoid unconsented manipulation.

Sources and limitations

Source register

Each entry states what it supports and what it cannot establish by itself. External links are visitor-initiated and send no referrer.

  1. AI-Enabled Emotional and Behavioral Manipulation: A Comprehensive Research Report

    Submitted research report retained in the private 2IA source corpus

    Supports
    Definitions, affective-computing critique, adaptive mechanisms, vulnerable populations, safeguards, and research gaps.
    Limit
    Some causal and legal claims in the report are broader than the underlying evidence; this adaptation uses restrained language.

    Preserved as private source evidence; no public file path is exposed.

  2. Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements

    Psychological Science in the Public Interest

    Supports
    Scientific limits of inferring emotion from facial movement.
    Limit
    Does not rule on every possible contextual or multimodal inference system.
    Open source
  3. Experimental evidence of massive-scale emotional contagion through social networks

    PNAS

    Supports
    Measured changes in emotional expression following feed manipulation.
    Limit
    Small effects on platform expression do not establish clinical harm or durable offline behavior.
    Open source
  4. Characterizing Manipulation from AI Systems

    Carroll et al.

    Supports
    Conceptual dimensions for identifying manipulation by AI systems.
    Limit
    A conceptual framework must be applied carefully to specific products and evidence.
    Open source