Resolution records
Every selected reference has a canonical title, publisher or venue, stable identifier, URL, category links, and reviewer state.
Source resolution · 91 records · automated checks are not human acceptance
Source resolution and bounded review
Every selected bibliography record now has a stable resolution entry, category links, claim links, source-type metadata, limitations, and an explicit reviewer state.
Resolution is not re-verification.
30 high-impact records received an automated retrieval or authoritative-source substitution. 61 remain metadata-resolved but independently unchecked. No named human reviewer has accepted the taxonomy.
Evidence boundary
What the release establishes—and what it does not
The twelve supplied reports remain the primary synthesis sources. All 91 selected bibliography records have resolution entries and are referenced by the 501-claim matrix and 4,157-edge citation graph. Thirty high-impact sources received bounded automated retrieval or authoritative-source substitution. The default v100.0.46 package has no authorized review key and no imported signed human-review artifact; human editorial review, publication acceptance, and full-corpus independent re-verification remain false.
Automated source checks
Selected high-impact records were retrieved, corroborated, or replaced by a stronger primary or official source.
Checks still pending
These records remain report-inherited bibliography entries. Their metadata is visible, but their scope has not been independently checked here.
Linked public claims
Every substantive taxonomy statement maps to one or more selected references. Orphaned claims fail validation.
30 of 91 sources received this bounded automated source check. The complete taxonomy remains not independently reverified.
Complete source-resolution registry
Checked and unresolved records remain visible together.
A checked record means the release process retrieved the selected source or a stronger authoritative substitute and recorded bounded support and limitations. It is not peer review, legal advice, or publication acceptance.
8 records shown.
No source records match all selected filters.
Remove one or more filters or reset the evidence view.
| Source | State | Publisher / type | Categories and claims | Scope and limitations |
|---|---|---|---|---|
Governing Evolving Memory in LLM Agents: Risks, Mechanisms, and the Stability and Safety Governed Memory (SSGM) Framework - arXiv (opens in a new tab) autonomous-influence-agents-s1
|
Automated scope check completeResearch Preprint RetrievedHuman reviewed: no | arXivResearch PreprintResearch | Autonomous Influence Agents40 linked claims |
Support and limitationsBounded support
Limitations
|
[2503.01829] Persuade Me if You Can: A Framework for Evaluating Persuasion Effectiveness and Susceptibility Among Large Language Models - arXiv (opens in a new tab) autonomous-influence-agents-s13
|
Automated scope check completeResearch Preprint RetrievedHuman reviewed: no | arXivResearch PreprintResearch | Autonomous Influence Agents40 linked claims |
Support and limitationsBounded support
Limitations
|
[2504.10286] Characterizing LLM-driven Social Network: The Chirper.ai Case - arXiv (opens in a new tab) autonomous-influence-agents-s22
|
Automated scope check completeResearch Preprint RetrievedHuman reviewed: no | arXivResearch PreprintResearch | Autonomous Influence Agents41 linked claims |
Support and limitationsBounded support
Limitations
|
PRC-linked influence operations are targeting AI debates in the US | OpenAI (opens in a new tab) autonomous-influence-agents-s23
|
Independent check pendingNot Independently RetrievedHuman reviewed: no | OpenAIOfficial Or InstitutionalOfficial Or Institutional | Autonomous Influence Agents41 linked claims |
This selected reference has not received an independent source-scope review in this release. |
PRC-linked influence operations are targeting AI debates in the US - OpenAI (opens in a new tab) autonomous-influence-agents-s24
|
Independent check pendingNot Independently RetrievedHuman reviewed: no | cdn.openai.comOfficial Or InstitutionalOfficial Or Institutional | Autonomous Influence Agents41 linked claims |
This selected reference has not received an independent source-scope review in this release. |
Evaluating AI Agent Persuasion of Safety Monitors - NeurIPS 2026 (opens in a new tab) autonomous-influence-agents-s27
|
Independent check pendingNot Independently RetrievedHuman reviewed: no | neurips.ccSecondary Or TechnicalSecondary Or Technical | Autonomous Influence Agents40 linked claims |
This selected reference has not received an independent source-scope review in this release. |
EU Commission Publishes Guidelines on the Prohibited AI Practices under the AI Act (opens in a new tab) autonomous-influence-agents-s34
|
Independent check pendingNot Independently RetrievedHuman reviewed: no | orrick.comSecondary Or TechnicalSecondary Or Technical | Autonomous Influence Agents40 linked claims |
This selected reference has not received an independent source-scope review in this release. |
Persuading large language models to comply with objectionable requests - PNAS (opens in a new tab) autonomous-influence-agents-s4
|
Independent check pendingNot Independently RetrievedHuman reviewed: no | Proceedings of the National Academy of SciencesResearch SourceResearch | Autonomous Influence Agents40 linked claims |
This selected reference has not received an independent source-scope review in this release. |
Evidence-tension register
Disagreement and scope limits are preserved.
Each tension carries competing bounded statements, public wording, source links, and a null human-decision field. Nothing here silently resolves a contested issue.
TENSION-PERSUASION-01 · Publication Wording Defined Human Adjudication Pending
Bounded persuasion effects versus broad microtargeting claims
One evidence boundary
A preregistered short conversational study found a stronger agreement shift when GPT-4 had basic participant information.
Countervailing boundary
Other report material questions whether psychographic microtargeting reliably adds persuasive value over high-quality generic messages in broader settings.
Public wording: State the design, sample, treatment, and measured outcome. Do not generalize short-term agreement shifts into durable behavior change or population control.
Editorial handling: State the design, sample, treatment, and measured outcome. Do not generalize short-term agreement shifts into durable behavior change or population control.
Reviewer decision: pending
TENSION-AUTONOMY-01 · Publication Wording Defined Human Adjudication Pending
Demonstrated agent coordination versus durable strategic autonomy
One evidence boundary
Laboratory and synthetic-platform studies demonstrate multi-agent interaction, role division, and bounded persuasion behavior.
Countervailing boundary
Reliable multi-month strategic coherence, operational security, and independent infrastructure management remain poorly evidenced.
Public wording: Separate demonstrated components from an end-to-end autonomous campaign. Use emerging or prospective labels for the latter.
Editorial handling: Separate demonstrated components from an end-to-end autonomous campaign. Use emerging or prospective labels for the latter.
Reviewer decision: pending
TENSION-RECOMMENDATION-01 · Publication Wording Defined Human Adjudication Pending
Algorithmic amplification versus claims of universal radicalization
One evidence boundary
Ranking and recommendation systems demonstrably shape exposure, salience, and incentives.
Countervailing boundary
The report corpus also notes that broad causal claims about algorithmically radicalizing the median user remain contested and are confounded by user choice and homophily.
Public wording: Describe exposure and amplification separately from persuasion, identity change, or offline behavior.
Editorial handling: Describe exposure and amplification separately from persuasion, identity change, or offline behavior.
Reviewer decision: pending
TENSION-PROVENANCE-01 · Publication Wording Defined Human Adjudication Pending
Provenance value versus provenance limits
One evidence boundary
Content Credentials can provide cryptographically verifiable, tamper-evident provenance indicators.
Countervailing boundary
Credentials do not establish that depicted events are true, provenance can be incomplete, and metadata can be removed or separated from an asset.
Public wording: Present provenance as one defensive signal alongside source verification, fact-checking, and forensics. Absence of a credential is not proof of fakery.
Editorial handling: Present provenance as one defensive signal alongside source verification, fact-checking, and forensics. Absence of a credential is not proof of fakery.
Reviewer decision: pending
TENSION-REACH-01 · Publication Wording Defined Human Adjudication Pending
Production scale versus authentic audience effect
One evidence boundary
Generative AI can reduce production friction, expand language coverage, and support high-volume content variation.
Countervailing boundary
OpenAI's 2024 disruption report found no meaningful authentic breakout for the five observed operations.
Public wording: Report production, distribution, authentic reach, engagement, persuasion, and behavior as separate measures.
Editorial handling: Report production, distribution, authentic reach, engagement, persuasion, and behavior as separate measures.
Reviewer decision: pending
TENSION-LAW-01 · Publication Wording Defined Human Adjudication Pending
Legal safeguards are time- and jurisdiction-dependent
One evidence boundary
The reports identify existing enforcement actions and regulatory controls relevant to synthetic media, sensitive data, manipulation, and automated decision systems.
Countervailing boundary
Case posture, effective dates, definitions, constitutional constraints, and remedies vary across jurisdictions and change over time.
Public wording: Date-stamp legal statements, identify the jurisdiction and procedural posture, and require current primary-source recheck before publication acceptance.
Editorial handling: Date-stamp legal statements, identify the jurisdiction and procedural posture, and require current primary-source recheck before publication acceptance.
Reviewer decision: pending
TENSION-CAPABILITY-PREVALENCE-01 · Publication Wording Defined Human Adjudication Pending
Controlled capability versus real-world prevalence
One evidence boundary
Controlled environments and purpose-built platforms show that model-driven agents can coordinate, interact, and generate varied influence content.
Countervailing boundary
Public evidence for reliable, long-duration, fully autonomous influence operations in adversarial real-world environments remains limited.
Public wording: Describe demonstrated agent and network capabilities without presenting them as proof of widespread autonomous deployment.
Editorial handling: Keep capability, autonomy level, duration, environment, and public attribution separate in every example.
Reviewer decision: pending
TENSION-SOURCE-AUTHORITY-01 · Authoritative Source Substituted Human Adjudication Pending
Aggregator or social-post references versus primary research
One evidence boundary
The supplied report bibliography includes a social-platform post summarizing detector-bias claims.
Countervailing boundary
A peer-reviewed primary study provides the appropriate bounded evidence for the detector-bias claim and reports dataset- and tool-specific results.
Public wording: Use the primary study for the public claim while retaining the original bibliography URL only as provenance.
Editorial handling: Do not repeat headline percentages from a social post as universal detector performance.
Reviewer decision: pending
TENSION-DETECTION-BIAS-01 · Publication Wording Defined Human Adjudication Pending
Detection utility versus false-positive and population-bias risk
One evidence boundary
Automated detectors and behavioral signals can support triage when combined with provenance and network evidence.
Countervailing boundary
Text and behavior detectors can misclassify non-native, neurodivergent, assisted, or highly formal human communication and should not be treated as conclusive identity evidence.
Public wording: Present automated signals as suggestive and require multi-signal human review before consequential action.
Editorial handling: No single detector score may establish that an account or text is synthetic.
Reviewer decision: pending
TENSION-LEGAL-CURRENTNESS-01 · Human Legal Currentness Review Required
Legal framework summaries versus current jurisdiction and procedural posture
One evidence boundary
The reports identify statutes, regulations, enforcement actions, and litigation relevant to manipulative or synthetic AI systems.
Countervailing boundary
Effective dates, amendments, injunctions, appeals, definitions, and remedies vary by jurisdiction and can change after the report date.
Public wording: Describe legal material as jurisdiction-specific and time-sensitive, and avoid presenting pending or preliminary matters as final holdings.
Editorial handling: Require source-specific currentness review by an authorized human before publication acceptance.
Reviewer decision: pending
Next evidence gate
Human editorial acceptance remains separate and pending.
The pending packet requires a named reviewer, bounded scope, category-by-category decisions, citation and safety checks, timestamp, and detached signature or explicitly authorized equivalent. This release does not simulate that decision.