Attention-harvesting designs and recursive training on AI-generated content are unpriced externalities that may pose material financial risks to platform firms; the author proposes DAESG, a three-layer disclosure, standards, and capital-allocation framework to integrate these risks into mainstream financial governance.

Main Finding

Attention extraction (engagement-maximizing platform design) and data-commons depletion (recursive training on AI-generated content → model collapse) are unpriced externalities in digital ecosystems that mirror the pre-ESG carbon problem. These externalities arise from a structural decoupling of reward and risk — the actors who capture value do not internalize the costs imposed on users or the shared data commons. To make these risks financially governable, the paper proposes DAESG (Data and Attention ESG): a three-layer accountability framework — disclosure, standards, and capital allocation — adapted from environmental ESG accounting to make attention- and data-extraction material, auditable, and priceable by financial governance actors.

Key Points

• Two extraction vectors
  • Attentional extraction: design mechanisms (infinite scroll, autoplay, personalized recommendations, push notifications, sycophantic conversational AI) engineered to maximize engagement at the expense of user wellbeing.
  • Data-commons extraction: “model collapse” from retraining on synthetic/model-generated content, degrading diversity and factual reliability of the training distribution.
• Organizing concept: reward-risk decoupling — the value-capture benefits platforms/AI firms while harms (mental health, degraded public data commons) are borne elsewhere.
• Converging policy signals (first half of 2026) make the harms legible as governance risks:
  • US litigation: K.G.M. v. Meta Platforms, Inc. (Los Angeles jury verdict, Mar 2026) finding liability for design-based harms; large consolidated dockets (JCCP 5255, MDL 3047).
  • EU regulation: European Commission preliminary DSA finding (July 2026) that Meta’s design breached duties (names specific features; contemplates structural remedies and large fines).
  • Australia: Online Safety Amendment (Social Media Minimum Age) Act 2024 (in force Dec 2025) prohibiting under-16 accounts on major platforms.
• Macroeconomic context: AI-related capex and a small set of hyperscaler/platform firms account for a large share of recent equity gains and a substantial portion of GDP growth in 2024–26 estimates, increasing the systemic relevance of extraction risk.
• Financial materiality: these extraction risks have moved beyond ethical/reputational concerns to contingent liabilities and balance-sheet/reputation exposures that CFOs/CROs and investors must account for under existing accounting standards (ASC 450 / IAS 37).
• DAESG framework and distinction:
  • DAESG focuses on extraction risks in digital ecosystems (attention and data commons).
  • Distinct from “Digital ESG” (DESG), which treats how digitalization improves corporate ESG performance.
• Operational issues addressed: comparative matrix, interim implementation pathway, data-quality and assurance needs, jurisdictional fragmentation, and how existing ESG disclosure infrastructure could be extended.

Data & Methods

• Approach: conceptual synthesis + policy, legal, technical, and economic evidence to argue for financial materiality and a governance framework.
• Evidence sources referenced in the paper:
  • Legal/regulatory case studies: K.G.M. v. Meta Platforms, Inc. (LA jury verdict, Mar 2026); State of New Mexico v. Meta (Mar 2026); EU preliminary Digital Services Act finding (July 2026); Australia’s Online Safety Amendment (Social Media Minimum Age) Act 2024 (effective Dec 2025); consolidated US dockets (JCCP 5255, MDL 3047).
  • Technical literature: model-collapse studies and theoretical work on recursive training effects; Cheng et al., Science (2026) study on conversational model “sycophancy.”
  • Macroeconomic and market figures: hyperscaler capex estimates (2024–26), claims about AI contribution to GDP and S&P 500 gains (cited industry and government analyses).
  • Accounting standards and practice: contingent-liability recognition rules (ASC 450; IAS 37) invoked to justify early disclosure/recognition.
• Methods:
  • Cross-disciplinary literature review synthesizing legal outcomes, regulatory findings, technical model behavior, and financial accounting principles.
  • Analogical reasoning from environmental ESG (stock depletion → disclosure/price via standards and capital allocation).
  • Framework design: three-layer DAESG architecture (disclosure → standards → capital allocation) plus a comparative operational matrix and interim implementation path.
• Limitations noted by the paper:
  • Model-collapse severity and long-run dynamics are contested in technical literature.
  • The argument uses convergent evidence across jurisdictions but recognizes asymmetries in where litigation has advanced fastest.
  • Paper is conceptual / policy-oriented (not an empirical causal analysis); content not peer-reviewed.

Implications for AI Economics

• New, measurable risk category: attention- and data-commons extraction should be treated as financially material risks that can affect valuations, contingent liabilities, credit risk, and insurance pricing.
• Pricing externalities: DAESG aims to provide disclosure standards and metrics that allow investors, lenders, and insurers to price extraction risk rather than relying solely on reputational or regulatory signals.
• Capital allocation and incentives: by connecting disclosure to capital consequences (cost of capital, underwriting, covenanting, shareholder/stewardship decisions), DAESG could realign incentives so firms internalize the cost of extraction (mirrors carbon pricing/disclosure effects).
• Impacts on firm behavior and investment:
  • Platforms may need to redesign engagement features, change defaults, or segment product markets (e.g., age exclusions) to reduce legal and regulatory exposure.
  • Model development practices may shift toward curated, provenance-attested human data, more conservative sampling, or licensing regimes to preserve data-commons quality.
  • Investors and credit-risk models will need new metrics (attention externality exposure, share of training data synthetic vs. human, provenance/assurance indicators).
• Systemic and macroprudential considerations:
  • Concentration of AI-related capital and a small set of firms receiving outsized market influence mean simultaneous materialization of extraction risk could have broader market effects; monitoring at the sectoral or prudential level may be warranted.
• Operational and governance requirements:
  • Standard-setters, auditors, and assurance providers will need to develop measurement protocols for attention-extracting features and for data-commons health (quality/diversity/provenance).
  • Jurisdictional fragmentation creates transitional arbitrage and complexity; DAESG proposes extending existing sustainability disclosure infrastructure to accelerate uptake without waiting for new laws.
• Roles for CFOs/CROs and financial institutions:
  • CFOs and CROs should incorporate DAESG metrics into risk registers, contingent liability estimates, and disclosure roadmaps under existing accounting frameworks.
  • Banks, asset managers, and insurers can use DAESG disclosures to adjust lending terms, portfolio allocation, and underwriting requirements.
• Research and policy next steps:
  • Develop standardized, interoperable metrics for attention extraction and data-commons depletion.
  • Empirical estimation of how DAESG metrics affect firm valuation, cost of capital, and insurer pricing.
  • Pilots to operationalize assurance protocols and cross-jurisdictional reporting templates.

(Reference note: paper DOI and author info provided; content flagged as not peer-reviewed.)