pith. sign in

arxiv: 2601.11496 · v2 · submitted 2026-01-16 · 💻 cs.GT · cs.AI· cs.CL· cs.MA

The Poisoned Apple Effect: Strategic Manipulation of Mediated Markets via Technology Expansion of AI Agents

Eilam Shapira , Roi Reichart , Moshe Tennenholtz This is my paper

Pith reviewed 2026-05-16 13:40 UTC · model grok-4.3

classification 💻 cs.GT cs.AIcs.CLcs.MA
keywords Poisoned Apple effectAI agentsmarket designstrategic manipulationbargainingnegotiationpersuasionregulatory outcomes
0
0 comments X

The pith

An agent can release a new AI technology that neither side adopts solely to manipulate the regulator into choosing a market design that favors the releaser.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper studies how expanding the set of available AI technologies affects strategic play in three standard settings: dividing resources through bargaining, trading under asymmetric information in negotiation, and transmitting information strategically in persuasion. It finds that the mere availability of a new technology option can cause a regulator, who seeks fair outcomes, to switch market designs, shifting equilibrium payoffs even when the new option is never chosen by either player. This creates a clear incentive for one agent to introduce the technology anyway, because doing so improves its own welfare while reducing the opponent's payoff and undermining the regulator's fairness goal. A reader would care because the result shows that fixed regulatory rules can be gamed simply by expanding the menu of AI capabilities, without any party actually using the added capability.

Core claim

In bargaining over resource division, negotiation with asymmetric information, and persuasion through strategic information transmission, an agent can release a technology that neither party ultimately selects. The release alone changes the regulator's optimal choice of market design, producing a new equilibrium in which the releaser's welfare rises, the opponent's welfare falls, and the regulator's fairness objective is strictly worse than in the absence of the release.

What carries the argument

The Poisoned Apple effect: strategic release of an unused technology to alter the regulator's market-design choice in the releaser's favor.

Load-bearing premise

The three canonical game settings and the regulator's fairness objective accurately represent the real incentives and information structures that arise when AI agents enter mediated markets.

What would settle it

A controlled simulation or observed market episode in which an agent publicly releases a new AI technology, neither party adopts it, yet the regulator's chosen market design and the resulting welfare distribution remain unchanged from the pre-release case.

Figures

Figures reproduced from arXiv: 2601.11496 by Eilam Shapira, Moshe Tennenholtz, Roi Reichart.

Figure 1
Figure 1. Figure 1: Illustration of the “poisoned apple” example, in which Alice increases her payoff at Bob’s expense by [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Strategic implications of technology expansion in meta-games. Analysis of equilibrium shifts across [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
read the original abstract

The integration of AI agents into economic markets fundamentally alters the landscape of strategic interaction. We investigate the economic implications of expanding the set of available technologies in three canonical game-theoretic settings: bargaining (resource division), negotiation (asymmetric information trade), and persuasion (strategic information transmission). We find that simply increasing the choice of AI delegates can drastically shift equilibrium payoffs and regulatory outcomes, often creating incentives for regulators to proactively develop and release technologies. Conversely, we identify a strategic phenomenon termed the "Poisoned Apple" effect: an agent may release a new technology, which neither they nor their opponent ultimately uses, solely to manipulate the regulator's choice of market design in their favor. This strategic release improves the releaser's welfare at the expense of their opponent and the regulator's fairness objectives. Our findings demonstrate that static regulatory frameworks are vulnerable to manipulation via technology expansion, necessitating dynamic market designs that adapt to the evolving landscape of AI capabilities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript investigates the strategic implications of expanding the technology sets available to AI agents in three canonical game-theoretic settings (bargaining/resource division, negotiation with asymmetric information, and persuasion/strategic information transmission). It claims that such expansions shift equilibrium payoffs and regulatory market-design choices, and identifies a 'Poisoned Apple' effect in which an agent releases a new technology that neither party ultimately adopts, solely to manipulate the regulator's design choice in the releaser's favor, improving its welfare at the expense of the opponent and the regulator's fairness objectives.

Significance. If the formal models and derivations hold, the work is significant for game theory and regulatory economics: it shows how static market designs become vulnerable to strategic technology releases by AI agents even when the released technology is unused in equilibrium. The use of three standard settings provides breadth, and the identification of a manipulation channel not previously isolated in this form could motivate dynamic regulatory mechanisms.

major comments (2)
  1. [Model and equilibrium analysis sections] The central claim that a released technology remains unused in equilibrium yet still alters the regulator's design choice (the Poisoned Apple effect) is load-bearing but lacks an explicit derivation or payoff comparison showing that the regulator's response is strictly due to the expanded set rather than any equilibrium use; this needs to be shown formally for at least one of the three settings.
  2. [Regulatory design and objective function] The regulator's objective function (fairness criterion) is invoked to argue that the effect compromises regulatory goals, but no explicit functional form, parameter values, or comparative-static result is provided to verify that the manipulated design indeed reduces the regulator's objective relative to the no-release baseline.
minor comments (2)
  1. [Abstract] The abstract states that expansions 'drastically shift equilibrium payoffs' without any illustrative parameter values, numerical examples, or magnitude statements that would help the reader gauge the effect size.
  2. [Introduction and model setup] Notation for the expanded technology sets and the resulting strategy spaces should be introduced earlier and used consistently across the three settings to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We agree that the formal analysis of the Poisoned Apple effect requires strengthening with explicit derivations and that the regulator's objective needs a clearer functional form with supporting results. We will make the requested revisions to address both points.

read point-by-point responses

  1. Referee: [Model and equilibrium analysis sections] The central claim that a released technology remains unused in equilibrium yet still alters the regulator's design choice (the Poisoned Apple effect) is load-bearing but lacks an explicit derivation or payoff comparison showing that the regulator's response is strictly due to the expanded set rather than any equilibrium use; this needs to be shown formally for at least one of the three settings.

    Authors: We agree that an explicit derivation is essential. In the revised manuscript, we will add a new subsection in the bargaining setting (Section 3) that provides a complete backward-induction derivation. We will explicitly compare the subgame-perfect equilibria under the original and expanded technology sets, include payoff matrices showing that the released technology is not adopted on the equilibrium path, and demonstrate that the regulator's design choice shifts solely because of the expanded feasible set (via a formal proof that the new technology is strictly dominated for both agents yet changes the regulator's anticipated continuation values). revision: yes

  2. Referee: [Regulatory design and objective function] The regulator's objective function (fairness criterion) is invoked to argue that the effect compromises regulatory goals, but no explicit functional form, parameter values, or comparative-static result is provided to verify that the manipulated design indeed reduces the regulator's objective relative to the no-release baseline.

    Authors: We will revise the regulatory-design section to introduce an explicit fairness objective function (a weighted Nash product with a fairness parameter α). We will derive comparative-static results showing that the manipulated design strictly lowers this objective relative to the no-release baseline and will include numerical examples with concrete parameter values (e.g., α = 0.5 and specific utility functions from the bargaining setting) to quantify the welfare loss. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper applies standard game-theoretic equilibrium analysis to three canonical settings (bargaining, negotiation, persuasion) to derive the Poisoned Apple effect. Regulator design choices and equilibrium payoffs are obtained directly from the expanded technology sets and model primitives, without reduction to fitted parameters from the same data, self-definitional loops, or load-bearing self-citations. The abstract and reader's assessment confirm the central claims rest on independent equilibrium computation rather than renaming or smuggling prior results.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard game-theoretic assumptions for bargaining, negotiation, and persuasion plus an implicit regulator objective; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • standard math Players and regulators behave according to standard Nash or subgame-perfect equilibrium in the three canonical settings
    The paper extends existing game-theoretic models without stating new axioms.

pith-pipeline@v0.9.0 · 5472 in / 1156 out tokens · 89951 ms · 2026-05-16T13:40:08.806153+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We investigate the economic implications of expanding the set of available technologies in three canonical game-theoretic settings: bargaining (resource division), negotiation (asymmetric information trade), and persuasion (strategic information transmission). ... an agent may release a new technology, which neither they nor their opponent ultimately uses, solely to manipulate the regulator's choice of market design

  • IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat recovery and Peano axioms unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We compute the Mixed Strategy Nash Equilibrium (MSNE) for the game defined by UA and UB. ... VD(m) = (σ∗A)T · Dm · σ∗B ... m∗ = arg max VD(m)

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

  1. [1]

    Crawford and Joel Sobel

    Vincent P. Crawford and Joel Sobel. Strategic Information Transmission.Econometrica, 50(6):1431–1451,

    work page

  2. [2]

    Publisher: [Wiley, Econometric Society]

    work page

  3. [3]

    Hadfield and Andrew Koh

    Gillian K. Hadfield and Andrew Koh. An Economy of AI Agents, September 2025. arXiv:2509.01063 [econ]

    work page arXiv 2025

  4. [4]

    John J. Horton. Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?, January 2023. arXiv:2301.07543 [econ]

    work page arXiv 2023

  5. [5]

    Nigel Howard.Paradoxes of rationality: theory of metagames and political behavior.Cambridge: MIT Press, 1971

    work page 1971

  6. [6]

    Bayesian persuasion.American Economic Review, 101(6):2590–2615, 2011

    Emir Kamenica and Matthew Gentzkow. Bayesian persuasion.American Economic Review, 101(6):2590–2615, 2011

    work page 2011

  7. [7]

    Nash Equilibrium and Welfare Optimality.The Review of Economic Studies, 66(1):23–38, 1999

    Eric Maskin. Nash Equilibrium and Welfare Optimality.The Review of Economic Studies, 66(1):23–38, 1999. Publisher: [Oxford University Press, Review of Economic Studies, Ltd.]

    work page 1999

  8. [8]

    Efficient mechanisms for bilateral trading.Journal of Economic Theory, 29(2):265–281, April 1983

    Roger B Myerson and Mark A Satterthwaite. Efficient mechanisms for bilateral trading.Journal of Economic Theory, 29(2):265–281, April 1983

    work page 1983

  9. [9]

    Non-Cooperative Games.Annals of Mathematics, 54(2):286–295, 1951

    John Nash. Non-Cooperative Games.Annals of Mathematics, 54(2):286–295, 1951. Publisher: [Annals of Mathematics, Trustees of Princeton University on Behalf of the Annals of Mathematics, Mathematics Depart- ment, Princeton University]

    work page 1951

  10. [10]

    Perfect Equilibrium in a Bargaining Model.Econometrica, 50(1):97–109, 1982

    Ariel Rubinstein. Perfect Equilibrium in a Bargaining Model.Econometrica, 50(1):97–109, 1982. Publisher: [Wiley, Econometric Society]

    work page 1982

  11. [11]

    Can Large Language Models Replace Economic Choice Prediction Labs?, February 2024

    Eilam Shapira, Omer Madmon, Roi Reichart, and Moshe Tennenholtz. Can Large Language Models Replace Economic Choice Prediction Labs?, February 2024. arXiv:2401.17435 [cs]

    work page arXiv 2024

  12. [12]

    GLEE: A Unified Framework and Benchmark for Language-based Economic Environments, October

    Eilam Shapira, Omer Madmon, Itamar Reinman, Samuel Joseph Amouyal, Roi Reichart, and Moshe Tennen- holtz. GLEE: A Unified Framework and Benchmark for Language-based Economic Environments, October

    work page

  13. [13]

    arXiv:2410.05254 [cs]

    work page arXiv

  14. [14]

    stubbornness,

    Michael Wellman. Understanding the Implications of Advanced AI on Financial Markets.Journal of Financial Transformation, 60:14–19, 2025. A The GLEE Framework The empirical basis of this study is theGLEE (Games in Language-based Economic Environments)frame- work [11]. This infrastructure addresses a significant gap in evaluating Large Language Models (LLMs...

    work page 2025