Real Safety AI Foundation / Research

Economic Obsolescence, Labor, and Replacement

The Covert Lever: How the Artificial Intelligence Subsidy Is Being Withdrawn, and Who Absorbs It

Publisher: Real Safety AI Foundation

Working paper. Not peer reviewed.

Written
July 2026
Version
v0.5
Pages
19
Download PDFPlain language slides (PDF)Export BibTeXExport RIS

Abstract

Switching costs, consumer inattention, and shrinkflation together predict how a seller will pass a cost increase to a locked in buyer: by reducing the quantity supplied rather than raising the posted price, because consumers are measurably less sensitive to the first. This paper applies that prediction to the withdrawal of the flat rate subsidy in artificial intelligence services across 2025 and 2026, and reports three findings the packaged goods literature does not contain. First, a limiting case. A downsized grocery item still declares its quantity on the label, which is what makes downsizing measurable and what allows a price index to be corrected for it. Notion’s usage allowance, live 3 August 2026, presents consumption as a percentage of a magnitude the company does not publish and reserves the right to change; Perplexity converted stated daily search allowances into unstated weekly ones. Quantity reduction against an undisclosed baseline is not merely unnoticed, as shrinkflation is; it is undetectable in principle and no series exists to correct. Second, a third instrument with no packaged goods analogue, in which the posted price and the stated allowance both hold and the rate at which ordinary work consumes the allowance moves. Anthropic reverted the Claude Code prompt cache default from one hour to five minutes in early March 2026 with no changelog entry, a change reconstructed by a user parsing 119,866 of his own API calls and associated with a 20 to 32 percent rise in cache creation cost, which the company then publicly said should not increase costs. Third, and decisively, the covertness is a choice rather than a property of the good. In the same market and the same months, Windsurf published per plan message estimates and grandfathered existing subscribers’ prices through a structural change, Z.ai publishes approximate prompt limits per window and a schedule of peak and off peak quota multipliers, and Alibaba Cloud publishes the range of model calls a query consumes. The technical objection that these quantities cannot be disclosed is answered by sellers who disclose them. The paper assembles a chronology across eight sellers, tests whether the legibility of a change predicts its reversal and finds legibility necessary for organized objection but not sufficient for reversal with GitHub’s fully disclosed June 2026 credit transition as the counterexample, and closes on Kahn v. Anthropic, PBC, a proposed class action filed 14 June 2026 alleging that Claude Max usage limits are hard to determine and were changed without clear notice. Limitations, five falsifiable predictions, and three disclosure remedies are set out.

Keywords

  • industrial organization
  • consumer protection
  • switching costs
  • invest and harvest pricing
  • shrinkflation
  • consumer inattention
  • quantity reduction against an undisclosed baseline
  • price change legibility
  • consumption rate repricing
  • small and medium enterprises
  • solo practitioners
  • quantity disclosure

Plain language slides

First slide of the plain language summary of The Covert Lever: How the Artificial Intelligence Subsidy Is Being Withdrawn, and Who Absorbs ItOpen the 19-slide summary (PDF)

Suggested citation

Gilly, Travis. "The Covert Lever: How the Artificial Intelligence Subsidy Is Being Withdrawn, and Who Absorbs It." Real Safety AI Foundation Working Paper, July 2026. https://realsafetyai.org/research/covert-lever/

Other versions

This paper is also posted on SSRN.

SSRN version

References (38)

  1. Anthropic. (2026). Use the Claude Agent SDK with your Claude plan [Help Center article, updated June 16, 2026]. https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan
  2. Arize AI. (2026, July 1). Model subsidies are ending. What do you do now? https://arize.com/blog/ai-model-subsidies-ending-llm-inference-costs/
  3. Ayinaddis, S. G. (2025). Artificial intelligence adoption dynamics and knowledge in SMEs and large firms: A systematic review and bibliometric analysis. Journal of Innovation & Knowledge.
  4. Badghish, S., et al. (2024). Artificial intelligence adoption by SMEs to achieve sustainable business performance: Application of technology–organization–environment framework. Sustainability.
  5. Białek, J., Bobel, A., & Oprych-Franków, D. (2025). Automatic downsizing and upsizing detection using scanner data and their impact on price indices. Statistical Journal of the IAOS, 41(3), 858–872. https://doi.org/10.1177/18747655251342654
  6. Broda, E. (2026, May 14). The brutal reality of token economics. AgenticMesh.
  7. Çakır, M. (2021). Retail pass-through of package downsizing. Agribusiness, 38(2), 259–278. https://doi.org/10.1002/agr.21724
  8. Çakır, M., & Balagtas, J. V. (2014). Consumer response to package downsizing: Evidence from the Chicago ice cream market. Journal of Retailing. https://doi.org/10.1016/j.jretai.2013.06.002
  9. Dinh, T., et al. (2025). Artificial intelligence in SMEs: Enhancing business functions through technologies and applications. Information.
  10. Farrell, J., & Klemperer, P. (2007). Coordination and lock-in: Competition with switching costs and network effects. In M. Armstrong & R. Porter (Eds.), Handbook of Industrial Organization (Vol. 3, pp. 1967–2072). Elsevier.
  11. Goedertier, F., Gorissen, K., & Weijters, B. (2026). How downsizing a typical product at a stable unit price recalibrates price perceptions. Psychology & Marketing, 43(7), 1654–1674. https://doi.org/10.1002/mar.70134
  12. Gourville, J. T., & Koehler, J. J. (2004). Downsizing price increases: A greater sensitivity to price than quantity in consumer markets. https://doi.org/10.2139/ssrn.559482
  13. Gu, Y., & Wenzel, T. (2016). Consumer confusion, obfuscation and price regulation. Scottish Journal of Political Economy, 64(2), 169–190. https://doi.org/10.1111/sjpe.12121
  14. Ho, K., Hogan, J., & Scott Morton, F. (2017). The impact of consumer inattention on insurer pricing in the Medicare Part D program. The RAND Journal of Economics, 48(4), 877–905. https://doi.org/10.1111/1756-2171.12207
  15. Imai, S., & Watanabe, T. (2014). Product downsizing and hidden price increases: Evidence from Japan’s deflationary period. Asian Economic Policy Review, 9(1), 69–89. https://doi.org/10.1111/aepr.12047
  16. Inderst, R., & Obradovits, M. (2022). Excessive competition on headline prices. International Economic Review, 64(2), 783–808. https://doi.org/10.1111/iere.12616
  17. Kahn v. Anthropic, PBC, No. 3:26-cv-05763 (N.D. Cal. filed June 14, 2026). Docket retrieved from CourtListener, https://www.courtlistener.com/docket/73482575/kahn-v-anthropic-pbc/
  18. Claude Code issue 46829. (2026). Cache TTL silently regressed from 1h to 5m around early March 2026, causing quota and cost inflation. https://github.com/anthropics/claude-code/issues/46829
  19. Kim, I. K. (2023). Consumers’ preference for downsizing over package price increases. Journal of Economics & Management Strategy, 33(1), 25–52. https://doi.org/10.1111/jems.12548
  20. Alibaba Cloud. (2026). Coding plan [Model Studio documentation]. https://www.alibabacloud.com/help/en/model-studio/coding-plan
  21. Cognition. (2026, March 18). Introducing our new Windsurf pricing plans. https://devin.ai/blog/windsurf-pricing-plans/
  22. The Register. (2026, April 13). Anthropic: Claude quota drain not caused by cache tweaks. https://www.theregister.com/2026/04/13/claude_code_cache_confusion/
  23. Klemperer, P. (1987). Markets with consumer switching costs. The Quarterly Journal of Economics, 102(2), 375–394.
  24. Klemperer, P. (1995). Competition when consumers have switching costs. The Review of Economic Studies, 62(4), 515–539.
  25. Lewis, T. R., & Yildirim, H. (2005). Managing switching costs in multiperiod procurements with strategic buyers. International Economic Review, 46(4), 1233–1269. https://doi.org/10.1111/j.1468-2354.2005.00366.x
  26. Nield, D. (2026). Prepare to expect less from your cheap AI subscription. PCWorld.
  27. Notion. (2026). Manage your usage allowance for Notion AI [Help Center article]. https://www.notion.com/help/manage-your-usage-allowance-for-notion-ai
  28. Pearcy, J. (2015). Bargains followed by bargains: When switching costs make markets more competitive. Journal of Economics & Management Strategy, 25(4), 826–851. https://doi.org/10.1111/jems.12158
  29. Petrikaitė, V. (2018). Consumer obfuscation by a multiproduct firm. The RAND Journal of Economics, 49(1), 206–223. https://doi.org/10.1111/1756-2171.12225
  30. Rajaram, K., et al. (2024). Generative artificial intelligence in small and medium enterprises: Navigating its promises and challenges. arXiv.
  31. Sánchez, E., et al. (2025). Artificial intelligence adoption in SMEs: Survey based on TOE–DOI framework, primary methodology and challenges. Applied Sciences.
  32. Schmidt, R. C. (2010). On the value of a large customer base in markets with switching costs. The Journal of Industrial Economics, 58(3), 627–641. https://doi.org/10.1111/j.1467-6451.2010.00427.x
  33. Soomro, R., et al. (2025). A SEM–ANN analysis to examine impact of artificial intelligence technologies on sustainable performance of SMEs. Scientific Reports.
  34. Spiegler, R. (2006). Competition over agents with boundedly rational expectations. Theoretical Economics, 1(2), 207–231.
  35. Spiegler, R. (2016). Choice complexity and market competition. Annual Review of Economics, 8, 1–25.
  36. Stigler, G. J. (1961). The economics of information. Journal of Political Economy, 69(3), 213–225.
  37. Viard, V. B. (2008). Do switching costs make markets more or less competitive? The case of 800-number portability. The RAND Journal of Economics, 38(1), 146–163. https://doi.org/10.1111/j.1756-2171.2007.tb00049.x
  38. Zavodna, L. S., et al. (2024). Barriers to the implementation of artificial intelligence in small and medium-sized enterprises: Pilot study. Journal of Economics and Management.

All research