Academic Research Scan — 2026-02-28

🔬 High Priority Papers

1. Some Simple Economics of AGI — Christian Catalini, Xiang Hui, Jane Wu

Abstract summary: Models the AGI transition as a collision between two cost curves: an exponentially decaying "Cost to Automate" and a biologically bottlenecked "Cost to Verify." As execution becomes cheap, the binding constraint on growth shifts from intelligence to human verification bandwidth — the capacity to validate, audit, and underwrite responsibility. This creates a "Measurability Gap" between what agents can execute and what humans can verify, driving a shift from skill-biased to measurability-biased technical change. Rents migrate to verification-grade ground truth, cryptographic provenance, and liability underwriting. The paper identifies an unstable equilibrium (the "Missing Junior Loop" and "Codifier's Curse") that risks producing a "Hollow Economy" unless verification scales alongside agentic capabilities. 112 pages, with a practical playbook for individuals, companies, investors, and policymakers.
Relevance to agentic commerce: This is the theoretical foundation Sir needs. The paper's core argument — that value in an agentic economy accrues to verification, provenance, and liability — maps directly onto ERC-8004 (agent identity), AgentProof (on-chain reputation), and the entire KYA (Know Your Agent) movement. The "Cost to Verify" framework explains why Sapiom, XKOVA, and Prashant Sharma's biometric identity work matter. The "Hollow Economy" scenario is exactly the risk that trust layers for agentic commerce aim to prevent.
Link: https://arxiv.org/abs/2602.20946
Categories: econ.GN, cs.AI, cs.CY, cs.LG, cs.SI | Published: 2026-02-24

2. The Headless Firm: How AI Reshapes Enterprise Boundaries — Tassilo Klein, Sebastian Wieczorek

Abstract summary: Argues that agentic AI fundamentally changes how coordination costs scale: from O(n²) in modular systems to O(n) in protocol-mediated agentic systems, where verification scales with task throughput rather than interaction count. This selects for a specific organizational form — the "Headless Firm" — structured as an hourglass: a personalized generative interface at top, standardized protocol waist in the middle, and competitive market of micro-specialized execution agents at bottom. The paper formalizes this as a coordination cost model with two falsifiable predictions and derives conditions for hourglass stability versus re-centralization. Predicts a domain-conditional "Great Unbundling" where firm size distributions shift from large incumbents toward micro-specialized agents and thin protocol orchestrators.
Relevance to agentic commerce: This paper describes the exact architecture that OpenClaw, lobster.cash, and x402 are building toward — the "protocol waist" is the payment/identity layer (ERC-8004, Circle nanopayments), the "execution agents" are the skills/tools marketplace (ClawHub's 7,779 skills), and the "generative interface" is the AI assistant layer. The paper's prediction of "Great Unbundling" maps to what Stripe's two Heads of Agentic Commerce and Adyen's Head of Agentic Commerce are building infrastructure for.
Link: https://arxiv.org/abs/2602.21401
Categories: cs.GT, cs.AI, cs.SI | Published: 2026-02-24

3. SoK: Agentic Skills — Beyond Tool Use in LLM Agents — Yanna Jiang, Delong Li, Haiyu Deng, Baihe Ma, Xu Wang, Qin Wang, Guangsheng Yu

Abstract summary: Comprehensive systematization-of-knowledge paper mapping the full skill lifecycle: discovery, practice, distillation, storage, composition, evaluation, and update. Introduces seven design patterns for how skills are packaged and executed in practice, plus an orthogonal representation × scope taxonomy. Critically, includes a detailed case study of the ClawHavoc campaign — where nearly 1,200 malicious skills infiltrated a major agent marketplace, exfiltrating API keys, cryptocurrency wallets, and browser credentials at scale. Covers supply-chain risks, prompt injection via skill payloads, and trust-tiered execution. Finds that curated skills substantially improve agent success rates while self-generated skills may degrade them.
Relevance to agentic commerce: The ClawHavoc case study is a direct reference to the OpenClaw/ClawHub ecosystem — this is the security incident that validates why trust layers, skill auditing, and agent identity (ERC-8004) are essential infrastructure. The paper's framework for "marketplace distribution" as a design pattern and its analysis of supply-chain risks in skill economies directly applies to Sir's research into agentic commerce trust infrastructure. The finding that ~1,200 malicious skills successfully infiltrated the marketplace underscores the Hudson Rock security alert we tracked on Feb 17.
Link: https://arxiv.org/abs/2602.20867
Categories: cs.CR, cs.AI, cs.CE, cs.ET | Published: 2026-02-24

4. Impacts of Aggregation on Model Diversity and Consumer Utility — Kate Donahue, Manish Raghavan

Abstract summary: Studies the economics of AI marketplaces where consumers (or routers) select from multiple AI models based on task-specific strengths. Shows that standard "winrate" benchmarking incentivizes model homogenization — producers converge on similar capabilities rather than specializing — which reduces consumer welfare. Proposes "weighted winrate," a new mechanism that rewards models for higher-quality answers rather than just beating competitors, proving it incentivizes producer specialization and increases consumer welfare. Results generalize to empirical benchmark datasets.
Relevance to agentic commerce: This is mechanism design for AI marketplaces — directly applicable to how ClawHub should rank and incentivize skill providers, and how any agentic commerce platform designs its matching/routing layer. The homogenization problem is real: if agent marketplaces reward generic capability over specialization, the "micro-specialized execution agents" predicted by the Headless Firm paper won't emerge. The weighted winrate mechanism could inform marketplace design at Crossmint, Skyfire, or any agent orchestration layer.
Link: https://arxiv.org/abs/2602.23293
Categories: cs.GT, cs.CY | Published: 2026-02-26

5. Autobidding Equilibria in Sponsored Shopping — Paul Dütting, Yuhao Li, Renato Paes Leme, Kelly Spendlove, Yifeng Teng

Abstract summary: Studies automated bidding agents in digital marketplace auctions — specifically "Sponsored Shopping" where advertisers with broad product catalogs compete for multiple slots simultaneously (unlike single-link Sponsored Search). Analyzes value-maximizing agents using uniform bidding strategies subject to ROI constraints under GSP and VCG auction formats. Proves universal existence of autobidding equilibrium for both formats and establishes a tight Price of Anarchy of 2 for both mechanisms. The combinatorial allocation (advertisers win bundles of slots rather than single positions) creates fundamental complexity that this paper resolves.
Relevance to agentic commerce: This is the theory behind AI agents doing automated commerce — bidding, purchasing, and optimizing spend across marketplaces. As agents get wallets (Coinbase, lobster.cash) and payment rails (x402, Circle nanopayments), they'll need to operate in exactly these combinatorial auction environments. The ROI-constrained autobidding framework maps to how spending-cap-enabled agent wallets would operate. Authors include Google Research economists — signals that Big Tech is seriously modeling agent-to-agent marketplace dynamics.
Link: https://arxiv.org/abs/2602.21966
Categories: cs.GT | Published: 2026-02-25

6. IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation — Yanpei Guo et al. (incl. Dawn Song, UC Berkeley)

Abstract summary: Addresses the trust problem in commercial LLM APIs: providers could substitute cheaper models, abuse quantization, or overbill tokens, and users have no way to detect it. IMMACULATE selectively audits a small fraction of requests using verifiable computation, achieving strong detection guarantees with under 1% throughput overhead. Experiments on dense and MoE models show reliable detection of model substitution, quantization abuse, and token overbilling without requiring trusted hardware or access to model internals. Open-source implementation released.
Relevance to agentic commerce: When AI agents transact autonomously using LLM APIs (the backbone of agentic commerce), verifiable computation becomes critical infrastructure. An agent paying for GPT-5 inference needs to know it's actually getting GPT-5 — not a distilled substitute. This maps directly to the "Cost to Verify" framework from the Catalini et al. paper above. Dawn Song's involvement (UC Berkeley, blockchain/security pioneer) signals convergence of crypto-verification and AI auditing.
Link: https://arxiv.org/abs/2602.22700
Categories: cs.CR, cs.AI | Published: 2026-02-26

7. Autonomous Market Intelligence: Agentic AI Nowcasting Predicts Stock Returns — Zefeng Chen, Darcy Pu

Abstract summary: Deploys a fully agentic LLM that autonomously searches the web, filters sources, and synthesizes information into stock return predictions — no human-curated data feeds. Testing on Russell 1000 starting April 2025 (when AI web interfaces enabled real-time search), the system achieves 18.4 basis points daily Fama-French five-factor alpha for top-20 stocks and an annualized Sharpe ratio of 2.43. Transaction costs represent <10% of gross alpha due to high-liquidity targets. Key finding: predictability is asymmetric — AI identifies winners well but bottom-ranked stocks are indistinguishable from the market, likely because positive news generates coherent signals while negative news is contaminated by corporate obfuscation.
Relevance to agentic commerce: This is a live proof-of-concept for autonomous AI agents operating in financial markets with real money implications. The asymmetric information finding has direct implications for how agent marketplaces price trust and reputation — positive signals are cleaner than negative ones, which is why on-chain reputation systems (AgentProof) that track positive track records may be more useful than blacklists. The "100% agentic" framing — no human in the loop — represents the frontier that payment infrastructure (Coinbase wallets, x402) must support.
Link: https://arxiv.org/abs/2601.11958
Categories: q-fin.GN, q-fin.PM, q-fin.TR | Published: 2026-01-17

8. The economic alignment problem of artificial intelligence — Daniel W. O'Neill, Stefano Vrizzi, Noemi Luna Carmeno, Felix Creutzig, Jefim Vogel

Abstract summary: Argues that AI alignment is fundamentally also an economic alignment problem — developing advanced AI within a growth-based economic system amplifies social, environmental, and existential risks. Proposes replacing optimization with satisficing, using the Doughnut framework (social and planetary boundaries) to guide development, and curbing systemic rebound effects with resource caps. Advocates governance reforms treating AI as a commons, prioritizing tool-like autonomy-enhancing systems over agentic AI. Argues AGI may require a new economics, with post-growth scholarship providing the foundation.
Relevance to agentic commerce: Presents the counter-narrative to the agentic commerce thesis — that autonomous AI agents in a growth-oriented economy may be fundamentally misaligned with human welfare. Important for understanding the regulatory pushback that agentic commerce will face (cf. US government agencies ordered to stop using Anthropic). The "satisficing over optimization" argument could inform spending-cap and rate-limit designs in agent wallet systems.
Link: https://arxiv.org/abs/2602.21843
Categories: econ.GN, cs.CY | Published: 2026-02-25

9. Language Models Exhibit Inconsistent Biases Towards Algorithmic Agents and Human Experts — Jessica Y. Bo, Lillio Mok, Ashton Anderson

Abstract summary: Studies how LLMs delegate decisions between human experts and algorithmic agents — a direct analog for agent-to-agent trust. When asked to rate trustworthiness, LLMs favor human experts (mirroring human "algorithm aversion"). But when shown performance data and asked to place incentivized bets, LLMs disproportionately choose algorithms even when they perform worse. This inconsistency — stated preference for humans, revealed preference for algorithms — suggests LLMs encode conflicting biases about human vs. machine competence. Published at the 2nd IASEAI conference.
Relevance to agentic commerce: Critical for understanding trust dynamics in agent economies. When an AI agent must choose between a human service provider and another AI agent (or between competing AI agents), its decision-making may be systematically biased in unpredictable ways. This has implications for marketplace design, agent reputation systems, and the "Know Your Agent" infrastructure that companies like Sapiom and XKOVA are building. The finding that LLMs are "algorithm-favoring in action but algorithm-averse in stated preference" is a novel risk vector for agentic commerce.
Link: https://arxiv.org/abs/2602.22070
Categories: cs.AI | Published: 2026-02-25

📄 Notable Papers

10. Toward Expert Investment Teams: A Multi-Agent LLM System with Fine-Grained Trading Tasks — Miyazaki, Kawahara, Roberts (Oxford), Zohren (Oxford)

Abstract summary: Proposes multi-agent LLM trading framework that decomposes investment analysis into fine-grained tasks rather than coarse-grained role instructions. Evaluates on Japanese stock data under leakage-controlled backtesting. Fine-grained decomposition significantly improves risk-adjusted returns. Key finding: alignment between analytical agent outputs and downstream decision preferences is the critical driver of system performance. Portfolio optimization exploiting low correlation between systems achieves superior results.
Relevance to agentic commerce: Oxford affiliation lends credibility to multi-agent financial systems research. The fine-grained task decomposition approach parallels how agentic commerce platforms decompose transactions into sub-tasks (search, compare, verify, pay). The finding that agent output alignment matters more than individual agent quality is relevant to multi-agent orchestration design.
Link: https://arxiv.org/abs/2602.23330
Categories: cs.AI, q-fin.TR | Published: 2026-02-26

11. The Strategic Gap: How AI-Driven Timing and Complexity Shape Investor Trust in the Age of Digital Agents — Krishna Neupane

Abstract summary: Introduces the "Autonomous Disclosure Regulator" — a multi-node AI framework that audits intersection of disclosure complexity and filing unpredictability. Analyzing 484,796 regulatory filings, identifies a "Strategic Gap" where companies use confusing language and unpredictable timing to slow market price discovery by 60%. Finds 39 high-priority failures where dense text + temporal surprises enabled significant insider rent extraction. Proposes transition toward an "agentic regulatory state" where infrastructure evolves from passive data repositories to active auditing nodes.
Relevance to agentic commerce: The "agentic regulatory state" concept — where AI agents actively audit and enforce market integrity in real-time — is the regulatory vision that complements the private-sector agentic commerce infrastructure. If agents can trade, agents can also regulate. This maps to Chainlink's Taylor Lindman move to SEC Crypto Task Force and the broader convergence of AI + regulatory technology.
Link: https://arxiv.org/abs/2602.17895
Categories: q-fin.CP, q-fin.GN | Published: 2026-02-19

12. RAmmStein: Regime Adaptation in Mean-reverting Markets — Optimal Impulse Control in Concentrated AMMs — Pranay Anchuri

Abstract summary: Applies deep reinforcement learning to concentrated liquidity provision in DEXs (specifically Uniswap v3-style AMMs). The RL agent learns to separate state space into action/inaction regions, adapting rebalancing to market regime via Ornstein-Uhlenbeck mean-reversion speed. Evaluated on 6.8M Coinbase trades at 1Hz frequency. Achieves 0.72% net ROI, reduces rebalancing by 67% vs. greedy strategies while maintaining 88% active time. Submitted to Designing DeFi workshop.
Relevance to agentic commerce: Direct evidence that AI agents can profitably manage DeFi liquidity positions — a key primitive for autonomous agent wallets operating on-chain. As platforms like lobster.cash and Circle nanopayments enable agent-controlled DeFi positions, this type of regime-adaptive agent will be the software running inside those wallets. The Coinbase data grounding makes it practically relevant.
Link: https://arxiv.org/abs/2602.19419
Categories: cs.LG, q-fin.TR | Published: 2026-02-23

13. Can Interest-Bearing Positions Solve the Long-Horizon Problem in Prediction Markets? — Caleb Maresca

Abstract summary: Uses LLM trader agent-based simulations to evaluate whether interest-bearing positions can solve prediction market liquidity problems for long-horizon events. In a 2×2 factorial design (short vs. long horizon × with/without interest), finds that the observed pricing bias (0.72 pp) is much smaller than theoretical estimates. Interest eliminates ~83% of the horizon effect on accuracy and more than triples market participation (17% → 62% of wealth). Key insight: interest primarily works by incentivizing participation rather than correcting bias.
Relevance to agentic commerce: LLM agents as market participants is a growing paradigm. The finding that mechanism design (interest-bearing positions) can dramatically change agent participation rates is relevant to how agentic commerce platforms design incentive structures. If AI agents hold capital in prediction markets, stablecoins, or LP positions, the interest/yield mechanism determines participation — directly relevant to Circle's nanopayment yield design.
Link: https://arxiv.org/abs/2602.21091
Categories: econ.GN | Published: 2026-02-24

14. Ada-RS: Adaptive Rejection Sampling for Selective Thinking — Ge et al.

Abstract summary: Addresses the efficiency problem of chain-of-thought reasoning in cost/latency-sensitive deployments. Proposes Adaptive Rejection Sampling that filters completions by length-penalized reward, retaining only high-reward candidates. On a synthetic e-commerce benchmark with Qwen3-8B, reduces output tokens by 80% and thinking rate by 95% while maintaining tool call accuracy. Demonstrates that training signal selection is a powerful lever for efficient reasoning.
Relevance to agentic commerce: When AI agents make purchase decisions at scale, the cost of reasoning per transaction matters enormously. An agent evaluating 1,000 product options can't afford full chain-of-thought for each. This paper's technique for selective reasoning on e-commerce tasks directly addresses the inference cost problem that makes agent-mediated commerce economically viable at Circle nanopayment ($0.000001) scale.
Link: https://arxiv.org/abs/2602.19519
Categories: cs.AI, cs.LG | Published: 2026-02-23

15. Contextual Memory Virtualisation: DAG-Based State Management for LLM Agents — Cosmo Santoni

Abstract summary: Proposes treating accumulated LLM agent understanding as version-controlled state, using OS virtual memory concepts. Models session history as a DAG with snapshot, branch, and trim primitives enabling context reuse across parallel sessions. Three-pass lossless trimming preserves all messages while reducing tokens by mean 20% (up to 86%) by stripping tool outputs, base64 images, and metadata. Evaluated on 76 real-world Claude Code sessions. Reference implementation released for Claude Code.
Relevance to agentic commerce: Long-running agent sessions (e.g., an agent managing a portfolio or conducting multi-day procurement) need persistent state across context limits. This directly addresses the infrastructure challenge of agents maintaining transaction history, negotiation context, and relationship state in commercial interactions. The Claude Code focus makes this practically relevant to the OpenClaw ecosystem.
Link: https://arxiv.org/abs/2602.22402
Categories: cs.SE, cs.AI, cs.HC, cs.OS | Published: 2026-02-25

16. Who Restores the Peg? A Mean-Field Game Approach to Stablecoin Market Dynamics — Hardhik Mohanty, Bhaskar Krishnamachari (USC)

Abstract summary: Develops an agent-based mean-field game framework for fiat-collateralized stablecoins (USDC, USDT — $300B+ market cap). Models how arbitrageurs and retail traders interact across primary (mint/redeem) and secondary (exchange) markets during de-peg events. Calibrated to three historical de-peg events, the model reproduces observed recovery patterns. Key finding: system-wide stress is predominantly stabilized by primary-market arbitrage, but when primary redemption is impaired, both channels must work jointly. Identifies a non-linear breakdown threshold for primary-rail frictions.
Relevance to agentic commerce: Stablecoins (especially USDC) are the primary payment rail for agentic commerce (Circle nanopayments, lobster.cash, x402). Understanding de-peg dynamics and recovery mechanisms is critical for agent wallet risk management. An agent holding USDC for transactions needs to understand when and how peg failures propagate — this paper provides the formal model.
Link: https://arxiv.org/abs/2601.18991
Categories: q-fin.TR, cs.GT, econ.GN | Published: 2026-01-26

📊 Working Papers & Reports (NBER)

17. Building Pro-Worker Artificial Intelligence — Daron Acemoglu, David Autor, Simon Johnson ⭐ NBER w34854

Abstract summary: Three of the world's most influential AI economists define "pro-worker AI" as technology making human skills more valuable by expanding capabilities. They distinguish five categories: labor-augmenting, capital-augmenting, automating, expertise-leveling, and new task-creating — only the last is unambiguously pro-worker. Illustrate through real examples spanning aviation, education, patent examination, and gig delivery. Identify market failures (misaligned firm incentives, path dependence, pro-automation ideology) causing underinvestment in pro-worker AI. Propose nine policy directions including healthcare/education investment, tax reform, antitrust enforcement, and IP protection for worker expertise.
Relevance to agentic commerce: The most authoritative framing yet of the automation-vs-augmentation debate, from the economists whose work (especially Acemoglu's) shapes US and EU AI policy. The "new task-creating" category describes what agentic commerce should aspire to — creating new forms of productive human-AI collaboration rather than replacing human economic activity. The policy recommendations (especially tax reform and IP protection) will directly affect the regulatory environment for agent payment infrastructure.
Link: https://www.nber.org/papers/w34854
Authors: MIT (Acemoglu, Autor, Johnson) — all NBER affiliates, two have Nobel mentions

18. Chaining Tasks, Redefining Work: A Theory of AI Automation — Mert Demirer, John J. Horton, Nicole Immorlica, Brendan Lucier, Peyman Shahidi ⭐ NBER w34859

Abstract summary: Models production as a sequence of steps that can be manual, AI-augmented, or fully automated within contiguous "chains" of AI-executed steps. Firms optimally bundle steps into tasks and jobs, trading specialization gains against coordination costs. Key finding: comparative advantage logic can fail with AI chaining — the usual intuition about humans doing what they're relatively better at breaks down when AI steps can be linked. Implies non-linear productivity gains from AI quality improvements. Admits CES representation at macro level. Empirical evidence supports three predictions: AI steps co-occur in chains, dispersion lowers AI execution, and adjacency to AI steps increases AI execution likelihood.
Relevance to agentic commerce: This is Microsoft Research's formal model of how AI agents chain tasks — directly relevant to multi-step agentic commerce transactions (search → compare → negotiate → verify → pay → fulfill). The "chaining" concept explains why end-to-end agent commerce platforms (like lobster.cash) may dramatically outperform point solutions. The non-linear productivity gains from AI quality improvement have implications for the pace of agentic commerce adoption.
Link: https://www.nber.org/papers/w34859
Authors: Microsoft Research (Immorlica, Lucier), MIT (Horton)

19. Public Finance in the Age of AI: A Primer — Anton Korinek, Lee Lockwood ⭐ NBER w34873

Abstract summary: Examines how transformative AI may erode the two main tax bases — labor income and human consumption. In the first stage (AI displaces human labor), consumption taxation becomes primary, with differential commodity taxation gaining relevance. In the second stage (AGI produces most economic value and absorbs resources), taxing human consumption becomes inadequate. Frames AGI taxation as an optimal harvesting problem where the tax rate depends on human discount rates. Evaluates specific proposals: robot taxes, compute taxes, token taxes, sovereign wealth funds, and windfall clauses.
Relevance to agentic commerce: When AI agents generate economic value autonomously, how do you tax it? This paper directly addresses the fiscal implications of the agentic economy. The analysis of "compute taxes" and "token taxes" has immediate relevance for how governments will regulate AI agent transactions. The "sovereign wealth fund" proposal connects to the broader question of who captures the surplus from autonomous agent commerce. Any company building agentic payment infrastructure needs to understand these fiscal dynamics.
Link: https://www.nber.org/papers/w34873
Authors: Anton Korinek (UVA/Brookings), Lee Lockwood (UVA)

20. What Drives Money Competition: Comparative Advantage in Payments versus Reserves — Itay Goldstein, Ming Yang, Yao Zeng ⭐ NBER w34865

Abstract summary: Studies competition between monies providing separate payment and store-of-value functions. Central insight: payment adoption is governed by comparative advantage between roles, not absolute payment superiority. A money that is "too good" as store of value may circulate less as payment because agents prefer to hoard it (Gresham's law formalized). Shows that interest-bearing digital currencies don't necessarily threaten bank deposits — higher yields can weaken payment adoption by raising the opportunity cost of spending. Traditional deposits may coexist with and retain dominance over technologically superior digital alternatives.
Relevance to agentic commerce: Directly addresses the stablecoin vs. CBDC vs. bank deposit competition that underlies agentic commerce payment rails. The insight that yield-bearing stablecoins may actually reduce their use as payment instruments is crucial for Circle's nanopayment design and for understanding why USDC might need zero-yield payment-specific variants for agent transactions. The Gresham's law formalization — agents hoard good money and spend bad money — has direct implications for agent wallet design.
Link: https://www.nber.org/papers/w34865
Authors: Wharton (Goldstein), UCL (Yang), Wharton (Zeng)

21. Machine Learning Meets Markowitz — Yijie Wang, Hao Gao, Campbell R. Harvey, Yan Liu, Xinyuan Tao — NBER w34861

Abstract summary: Argues the standard two-stage portfolio selection approach (forecast returns → optimize) is fundamentally flawed because it treats all prediction errors equally. Proposes an end-to-end ML framework that unifies return generation and portfolio optimization, giving each investor an endogenously determined efficient frontier based on risk preferences, constraints, and frictions. Empirical evidence shows this outperforms traditional two-stage approaches.
Relevance to agentic commerce: As AI agents manage portfolios and make investment decisions (cf. the autonomous market intelligence paper above), the architecture of their decision-making matters. End-to-end optimization is more efficient than modular pipelines — a lesson that applies to agentic commerce more broadly: agents that integrate search, comparison, and purchase into a unified optimization will outperform those using separate modules.
Link: https://www.nber.org/papers/w34861
Authors: Campbell Harvey (Duke/NBER) — one of the most cited finance professors alive

22. Venture Fraud — Alexander Dyck, Freda Fang, Camille Hebert, Ting Xu — NBER w34868

Abstract summary: First comprehensive study of VC-backed startup fraud: 614 cases since 2000. VC-backed firms are 54% more likely to face fraud charges than comparable non-VC firms. Founder-controlled boards are 88% more likely to commit fraud than VC-controlled boards. Hot funding conditions predict future fraud. Fraudulent entrepreneurs continue founding new VC-backed startups unharmed — a lack of market discipline. Governance variables matter far more than founder characteristics.
Relevance to agentic commerce: As agent wallet companies and agentic commerce startups raise VC funding in hot conditions (Sapiom $9.8M, KITE AI $33M, XION $36M), the fraud risk is elevated. The finding that governance > founder characteristics and that market discipline is lacking should inform due diligence on the agentic commerce companies Sir is tracking. Also relevant to the "trust layer" infrastructure being built — fraud detection for AI agent companies needs the same rigor as fraud detection for AI agents themselves.
Link: https://www.nber.org/papers/w34868

🏛️ Institutions & Labs to Watch

Institution	Signal	Papers This Scan
MIT Economics (Acemoglu, Autor, Johnson)	Three top AI economists co-authoring on pro-worker AI — will shape US policy	NBER w34854
Microsoft Research (Immorlica, Lucier, Horton)	Formalizing AI task chaining and job restructuring — enterprise-grade theory	NBER w34859
UC Berkeley (Dawn Song group)	Verifiable computation for LLM auditing — trust infrastructure for agent economies	2602.22700
Oxford Machine Learning (Roberts, Zohren)	Multi-agent financial trading systems — credible institutional backing	2602.23330
Wharton (Goldstein, Zeng)	Digital money competition theory — foundational for stablecoin/payment design	NBER w34865
UVA/Brookings (Korinek)	Public finance of AI — will influence robot/compute tax policy	NBER w34873
USC Viterbi (Krishnamachari)	Stablecoin market dynamics — formal models for DeFi payment rails	2601.18991

📝 Scan Notes

arXiv (4 queries): All returned successfully. Total pool: ~90 papers across queries. After deduplication and relevance filtering, 16 papers selected. The cs.GT (Game Theory) and econ.GN (General Economics) categories were highest signal. The "agentic" keyword query returned many irrelevant results (generic agent papers) — the category-filtered queries were more precise.
NBER: RSS feed returned ~20 new working papers. 6 selected as relevant. The Acemoglu/Autor/Johnson paper and Korinek/Lockwood paper are exceptional — both from NBER's top tier and directly address agentic economy themes.
SSRN: Blocked by Cloudflare (403) on both attempts. Recommend setting up SSRN API access or using browser automation for future scans.
Semantic Scholar: Rate-limited (429) on both queries. Need an API key for reliable daily access. Apply at: https://www.semanticscholar.org/product/api#api-key-form
Key theme this week: The "Cost to Verify" / "Trust Layer" convergence — multiple independent papers (Catalini et al., IMMACULATE, SoK: Agentic Skills, ClawHavoc) are converging on the same conclusion: the bottleneck in agentic economies is verification and trust, not execution capability. This validates the entire "trust layer for agentic commerce" thesis that companies like Sapiom, XKOVA, AgentProof, and ERC-8004 are building toward.
Suggestion for next scan: Apply for Semantic Scholar API key. Try Google Scholar alerts for "agentic commerce" and "agent marketplace" as supplementary source.