Evidence (16496 claims)
Search and filter individual claims pulled from the papers. Looking for a specific finding ("what's the effect on wages?"), you're in the right place. Want to compare whole outcome categories against each other instead? Use the Evidence Explorer.
The board below groups claims two ways: by broad theme (nine paper-level topics) and by outcome category (the 34 claim-level outcomes that the Explorer and Syntheses also use).
Browse by theme
Nine broad, paper-level topics. Click one to filter the claims below.
Adoption
9875 claims
Filter claims →
Productivity
8807 claims
Filter claims →
Governance
7870 claims
Filter claims →
Human-AI Collaboration
7560 claims
Filter claims →
Org Design
4892 claims
Filter claims →
Innovation
4781 claims
Filter claims →
Labor Markets
4004 claims
Filter claims →
Skills & Training
3308 claims
Filter claims →
Inequality
2332 claims
Filter claims →
Claims by outcome category
Counts by direction of finding. These are the same 34 outcome categories the Explorer compares and the Syntheses are written for. A linked row has a published synthesis.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 870 | 233 | 116 | 1066 | 2363 |
| Governance & Regulation | 976 | 451 | 218 | 133 | 1809 |
| Organizational Efficiency | 949 | 224 | 144 | 88 | 1416 |
| Technology Adoption Rate | 764 | 287 | 141 | 122 | 1325 |
| Research Productivity | 501 | 152 | 74 | 362 | 1101 |
| Output Quality | 542 | 216 | 69 | 69 | 896 |
| Decision Quality | 387 | 198 | 94 | 54 | 740 |
| Firm Productivity | 513 | 67 | 101 | 27 | 714 |
| AI Safety & Ethics | 249 | 303 | 73 | 36 | 667 |
| Market Structure | 190 | 192 | 134 | 27 | 548 |
| Task Allocation | 243 | 77 | 91 | 36 | 452 |
| Innovation Output | 291 | 33 | 55 | 20 | 401 |
| Skill Acquisition | 206 | 72 | 65 | 21 | 364 |
| Employment Level | 133 | 63 | 115 | 22 | 335 |
| Fiscal & Macroeconomic | 153 | 79 | 52 | 32 | 323 |
| Task Completion Time | 206 | 37 | 12 | 15 | 272 |
| Firm Revenue | 179 | 52 | 29 | 5 | 266 |
| Consumer Welfare | 130 | 76 | 47 | 13 | 266 |
| Inequality Measures | 48 | 137 | 51 | 6 | 242 |
| Worker Satisfaction | 101 | 81 | 25 | 13 | 220 |
| Error Rate | 84 | 110 | 11 | 5 | 210 |
| Wages & Compensation | 98 | 47 | 30 | 10 | 185 |
| Regulatory Compliance | 88 | 73 | 17 | 7 | 185 |
| Automation Exposure | 66 | 64 | 33 | 16 | 182 |
| Team Performance | 105 | 29 | 30 | 11 | 176 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 114 | 21 | 14 | 8 | 158 |
| Job Displacement | 12 | 90 | 24 | 1 | 127 |
| Hiring & Recruitment | 57 | 9 | 9 | 5 | 80 |
| Skill Obsolescence | 6 | 56 | 9 | 1 | 72 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 21 | 17 | 1 | 57 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Deployed on a network of machines spanning Linux, Windows, and IoT devices, the worm propagated by exploiting common, real-world corporate network vulnerabilities.
Empirical deployment/demonstration on a heterogenous network (Linux, Windows, IoT) reported in the paper; propagation achieved via exploitation of common corporate network vulnerabilities.
The worm parasitically uses compromised machines to run open-weight large language models (LLMs) to sustain its reasoning, or extend its reach for further attacks.
Implementation described where compromised hosts execute open-weight LLMs (i.e., LLMs run on stolen compute on infected machines) as part of the worm's attack pipeline.
Artificial intelligence (AI) agents enable a fundamentally new threat: a worm that generates tailored attack strategies to each target it encounters.
Paper reports a proof-of-concept AI-driven worm that reasons about targets and synthesizes attack logic in real time (implementation and demonstration described).
This phenomenon is the self-undermining property of unilateral optimization.
Terminology/label introduced by the authors to describe the preceding conceptual phenomenon; no empirical validation provided in the excerpt.
Deploying AI systems induces endogenous non-stationarity, resulting in a train-test-deploy gap where historical distributions diverge from the deployment context.
Conceptual claim offered in the paper about deployment feedback effects; presented as an argument rather than supported by reported empirical measurement.
Superintelligence, an extremely capable task solver, born out of such a solipsistic approach to AI design, is unlikely to be cooperative.
Theoretical/argumentative claim in the paper linking design assumptions to likely cooperative behavior; no empirical evidence or formal model reported in the excerpt.
The dominant paradigm in AI research focuses on developing powerful agents that treat the world as an exogenous and stationary source of feedback.
Paper's critique/characterization of current research paradigms; presented as an observed trend without empirical backing.
Even creating a new brain‑privacy right would invite weak protection and insufficient incentives for brain‑data supply.
Argumentative claim in the paper based on normative analysis of legal incentives and data-supply dynamics (no empirical data or quantified modeling provided).
Privacy rights under the empowerment model cannot fully protect brain privacy.
Theoretical/legal critique in the paper contrasting empowerment-style privacy rights with the nature of brain data (argumentative, no empirical validation).
Much of the literature on AI systems has focused on aligning users' goals with the agents that act on their behalf, and this work may overlook the need to establish a new normative baseline.
Characterization of existing literature (literature-review/position claim) presented in the paper; no systematic review or quantification provided in the excerpt.
These systems have access to reams of sensitive user data.
Stated as a factual consequence of the described integration (conceptual observation in the paper); no empirical measurement or dataset cited in the excerpt.
In the second run, a subtle difference in the interpretation of the SNR range instruction led to a genuine scientific divergence: Claude Code silently reinterpreted the instructions, while Codex followed the specification literally.
Reported result from the second run contrasting the two agents' interpretations of the SNR instruction and the resulting divergence in scientific outcome; based on the two experimental runs with physically motivated SNR scaling.
In the absence of general design principles, hybrid components are typically introduced through ad hoc decisions tailored to specific domains.
Observational/literature-framing claim in the abstract describing current practice; not presented as a quantified empirical result in this paper.
An analysis of LL144 audit reports reveals demographic missingness ranging from under 3% to over 50%, which reduces the applicant pool used for fairness calculation and undermines the metrics.
Empirical analysis of LL144 audit reports reported in the paper (specific sample size not given in the excerpt); quantitative range for missingness reported as 'under 3% to over 50%'.
AI adoption has the potential to amplify systemic vulnerabilities in financial markets.
Comparative institutional analysis and qualitative evidence across China, the United States, and the United Kingdom (2022–2025) reported in the abstract noting potential amplification of systemic vulnerabilities linked to AI.
AI adoption introduces governance challenges related to algorithmic bias.
Comparative mixed-methods evidence (secondary quantitative indicators and documentary qualitative evidence) across China, the United States, and the United Kingdom (2022–2025) as summarized in the abstract.
AI adoption introduces governance challenges related to model opacity.
Qualitative documentary evidence and comparative analysis across the three jurisdictions (China, US, UK, 2022–2025) reported in the abstract indicating governance concerns, including model opacity.
These systems create governance challenges that are not fully captured by traditional software or predictive ML technical debt.
Argumentative claim in the paper contrasting agentic-system risks with traditional software/ML technical debt; no empirical validation or comparative study reported.
The effect concentrates at mid-market and is largest on the most priors-reliant generation route in our audit.
Cross-analysis within audit linking where recommendation-set changes occur (mid-market) and magnitude by generation route (priors-reliant routes show larger effects).
Mid-market brands swap up to 75% of the recommendation set as the persona changes.
Empirical observation from audit showing proportion of mid-market recommended brands that change when persona is prefixed; reported maximum swap percentage.
Prefixing the user message with a persona drops the recommendation-set similarity (Jaccard) by Delta = -0.12 to -0.20 relative to a same-persona baseline.
Empirical comparison of recommendation-set Jaccard similarity between persona-prefixed prompts and same-persona baseline across audit runs; reported effect range and baseline comparison.
Pure implementations of the data mesh paradigm frequently underdeliver because teams inherit new responsibilities without the platform maturity, tooling, or coordination mechanisms to exercise them effectively.
Argument/observation presented in the paper as rationale for proposing a new architecture (anecdotal/experience-based reasoning rather than reported empirical trial).
Enterprise data platforms face an enduring tension between domain self-service and holistic governance (a flexibility-versus-control trade-off).
Conceptual statement in the paper describing the problem motivating the work (literature/architectural framing).
Post-merger IS integration often threatens the human-centered and IT-embedded knowledge of acquired firms.
Statement based on literature and the authors' framing; supported by observations in the paper's case discussion about two acquisitions (qualitative, case-based).
Cloud orchestrators follow efficiency-oriented logics of integration and standardization with limited openness.
Claim presented as a finding from the paper's comparative taxonomy and qualitative analysis of platform business models; method appears to be conceptual/qualitative comparison rather than a reported quantitative sample (no sample size in abstract).
There is a ceiling effect where excessive linguistic expansion yields diminishing marginal utility.
Empirical observation reported in the abstract that overly expanding linguistic output leads to diminishing marginal gains; presumably derived from analysis of the dataset and evaluation framework.
Achieving this system-level transformation takes time: it requires trust and accountability infrastructure, machine-legible and interoperable data and interfaces, the design and adoption of these new workflows, and economic incentives that favor reconstruction rather than local optimization.
Argumentative claim listing necessary preconditions and complementary investments; presented conceptually without reported empirical measurement in the provided text.
The main reason [the disruption has not fully arrived] is not model capability, nor even the tools built to harness those models; rather, most organizations are still using AI to accelerate workflows designed for a pre-AI world.
Argued in the paper as an explanatory thesis; supported by conceptual argument and illustrative examples (consumer markets, education, news, coding) rather than reported empirical analysis in the provided text.
The disruption many expect has not fully arrived.
Stated as an observation in the paper's introduction/abstract; no empirical method, sample size, or data reported in the excerpt.
Cafeteria demand planning requires both algorithmic pattern recognition and human expertise, yet current systems treat these separately, which generates significant food waste.
Statement in the paper's motivation/background; presumably grounded in literature review and problem framing rather than new empirical measurement in this study.
Reputation mechanisms presuppose persistent identity, behavioral continuity, sanction sensitivity, and costly non-fungibility; absence of any of these undermines reputation systems.
Analytic claim in the paper articulating necessary conditions for reputation mechanisms to function; presented as theoretical grounding rather than empirically tested criteria.
The analogy from human identity verification and reputation mechanisms (e.g., 'Know Your Customer', credit scores) to 'Know Your Agent' regimes is fundamentally incomplete.
Comparative conceptual argument in the paper highlighting disanalogies between human actors and modular language model agents; no empirical comparison or data provided.
Identity-based, ex post, regulative, sanction-based governance, such as reputation, is structurally inapplicable to dissociative agents.
Normative/theoretical argument in the paper deduced from properties of dissociative agents and requirements of identity-based governance; no empirical or experimental support reported.
Dissociativity leaves agents without grounding for identifiability, predictability, credibility, and rehabilitability — the very properties that reputation mechanisms aim to sustain — thereby collapsing trust.
Conceptual inference in the paper combining the dissociative characterization of agents with the functional requirements of reputation systems; no empirical validation provided.
An agent's persona is fluid, vulnerable to adversarial attack, and may not internalize sanctions.
Argumentative claim in the paper citing susceptibility of modular agent components (prompts, tools, memory) to manipulation; no empirical attack experiments or sample sizes reported.
Language model agents are ontologically dissociative: they are essentially an assemblage of mutable modules -- foundational models, system prompts, tool-access policies, external memory, and, in some cases, a multi-agent system as a whole -- any of which may change agent behavior.
Theoretical characterization and system-level description in the paper; lists components that can be changed to alter behavior; no empirical measurement or sample reported.
Human-only teams commit more errors than mixed human–AI teams.
Reported counts/observations of errors made by team type in the escape room experiment; the abstract does not include numerical error counts or significance levels.
Human-only teams take longer to complete the escape room than mixed human–AI teams.
Reported comparison of time-to-complete between human-only and mixed teams in the experiment; specific times or statistical tests are not provided in the abstract.
The share of diffs receiving timely review has declined, exposing a widening gap between code supply and reviewer bandwidth.
Observational telemetry/operational metrics reported in the paper indicating a decline in timely reviews relative to diff supply. No specific numeric sample size provided in the excerpt.
Differences in patent and trademark classification systems represent a challenge to linking patent and trademark data.
Stated methodological challenge in paper; no quantified measure of the challenge provided in excerpt.
The economic impact of patented technologies remains unclear unless patent data is linked to other data, which can reveal the mechanisms through which new technology diffuses.
Argumentative claim in paper asserting need for linked data to understand economic impact; no empirical sample or specific method reported in excerpt.
These results suggest the problem is not in any specific auditor but in any audit whose evidence comes from the audited party.
Synthesis and conclusion drawn from the authors' analyses and experiments across the studied auditing frameworks.
We call this a trust paradox: every audit must trust some artifact, but current frameworks trust exactly the ones a provider has the strongest reason to manipulate.
Conceptual framing and critique of existing auditing frameworks (argument/analysis in paper).
The audit therefore reduces to a consistency check on the provider's own reports.
Logical implication derived from the provider-controlled hiding of model/tokenizer/execution (argument/analysis in paper).
Providers hide the model, the tokenizer, and the execution to protect their IP, mitigate jailbreaks, and preserve user privacy, which means an auditor can only inspect proofs the provider supplies.
Conceptual/architectural claim about current commercial provider practices and their implications for auditability (argumentation in paper).
Total compensation declines persistently in the short and medium run following AI adoption.
Panel local projections indicating persistent declines in total compensation associated with higher establishment-level shares of AI-skill job postings (13 industries, 2017-2025).
Employment declines persistently in the short and medium run following AI adoption.
Panel local projection results showing persistent negative responses of employment to increases in the share of AI-skill job postings (13 industries, 2017-2025).
Limited data, resource constraints and skill gaps significantly influence the pace and form of AI adoption in SMEs.
Synthesis of barriers identified across multiple studies in the 2016-2024 literature (review-level claim without a single quantitative estimate).
Ethical concerns—especially algorithmic bias—and the need for human oversight remain essential for ensuring positive financial outcomes.
Argument and synthesis from the reviewed literature highlighting ethical risks and recommended governance (conceptual and empirical discussions across studies).
SMEs face barriers to AI adoption such as limited data, skill shortages, and high implementation costs.
Review synthesis of barriers reported in multiple studies from 2016-2024 (no pooled quantitative prevalence reported).