Pith. sign in

REVIEW 2 major objections 4 minor 38 references

AI-Native Insurance for Agentic AI: Pricing, Underwriting, and End-to-End Automation

T0 review · 2 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read One formula decides whether an AI agent deployment is insurable.

desk verdict Solid framework for agentic-AI insurance, but the fixed-terms insurability theorem skips the incentive-compatibility constraint it needs; the gap is fixable but the practical pricing claims outrun the evidence. read the letter →

arxiv 2607.13230 v1 pith:T6I46NEJ submitted 2026-07-14 cs.AI cs.CRcs.CY

Quanyan Zhu This is my paper
classification cs.AIcs.CRcs.CY MSC 91B3091B41
keywords agenticAIinsurancecyberrisk-stateunderwritingcontractdesigngovernancecertificationinsurabilityregionrisk-basedpricing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that whether an agentic-AI deployment can be insured at fixed coverage terms is decided by a single closed-form quantity: the marketable surplus Δ_i(s_i) = J_i^0 − K_i(g_i) − Σ_e q_i^e x_i^e − ρ_i(s_i,C_i). A deployment is insurable exactly when governance cost plus expected gross loss plus risk loading stays below the uninsured baseline minus a buffer. From this identity the paper derives three structural results: insurability is a region of the risk-state space, feasibility worsens monotonically as exposure grows, and a governance tier can certify a deployment as insurable. A healthcare case study shows the full pipeline—pricing, coverage allocation, governance covenants, and automated claims—as one constrained optimization. If the framework is right, insurers could price agentic-AI risk from underwritable observables rather than case-by-case judgment.

What carries the argument

The risk state s_i=(α_i,β_i,η_i,g_i,v_i) reduces an agentic deployment to five underwritable coordinates: autonomy category, operational authority, permission vector, governance tier, and dependency shares. Mappings Q^e(s_i) and X^e(s_i) convert these state coordinates into event probabilities and severities; the coverage-incidence matrix Γ_i and allocation matrix Λ_i separate which coverage layers can respond to an event from how payment is divided among them. The surplus identity (20) is the load-bearing object: it collapses the entire feasibility question into a comparison of four scalars, and the governance-certification result converts governance from a qualitative virtue into a contrac

What would settle it

Find two agentic deployments s ⪯_E s′ with the same governance tier where s′ is more exposed in authority, permissions, or dependency concentration, yet observed annual loss frequency or severity is lower for s′ than for s. Proposition 5 requires the marketable surplus to be nonincreasing in exposure, so such a pair would falsify the monotone-deterioration claim. Equivalently, an empirical estimate of Φ(g)=K(g)+Σ_e q^e(s) x^e(s)+ρ(s,C) that increases when governance improves from tier g to g′ would invalidate the governance-certification threshold of Proposition 6.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the indemnity schedule—deductibles, limits, and allocation shares—only determines where an acceptable premium lies, not whether one exists. Marketability reduces to identity (20): the maximum acceptable premium minus the risk-loaded minimum premium equals J_i^0 − K_i(g_i) − Σ_e q_i^e x_i^e − ρ_i(s_i,C_i), with the contract's indemnity terms canceling out. Insurability therefore depends only on the uninsured baseline, governance cost, expected gross loss, and risk loading. Consequently, under exposure-monotone probability, severity, and loading maps, insurability is downward closed in exposure, and if stronger governance is net risk-reducing, there is a m

Load-bearing premise

The structural results collapse if the probability, severity, and risk-loading maps are not monotone in exposure (and, for the governance threshold, if stronger governance is not net risk-reducing), because these maps are posited as benchmark specifications calibrated by scorecards and expert elicitation rather than estimated from agentic-AI claims data.

Editorial extensions

If this is right

  • Underwriting an agentic-AI deployment can proceed by estimating five observable state coordinates and four scalar cost components; the premium then lies in a closed interval [T_min, T_max−s_0].
  • Any one-factor increase in delegated authority, permission exposure, or dependency concentration can only shrink or close the feasible premium interval; it can never restore insurability along the same governance tier.
  • Stronger governance, when it reduces total expected risk cost, has a threshold effect: below the certifying tier no premium clears the market; at or above it a mutually acceptable contract exists.
  • Insurance can act as an operating cost and regulatory mechanism: bundled or mandated premiums behave like a risk price that screens deployments, with the risk state giving regulators a target for financial-responsibility mandates.
  • The same contract can be executed online: monitoring, trigger evaluation, claim validation, and settlement can be automated, with human review reserved for ambiguous, fraudulent, or catastrophic exceptions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • One testable extension: collect claims and telemetry data and check whether the empirical frequency-severity product is actually monotone in authority, permissions, and dependency concentration; a single exposure-monotonicity violation would reshape the insurability region.
  • The governance-certification result implies that insurers could publish tier-based insurability certificates, but only if governance evidence maps consistently to realized risk reduction; otherwise certification could reward box-checking.
  • The paper treats risk loading per policy, yet dependency concentration R(v_i) hints at systemic correlation across insureds sharing one model provider; extending the framework to portfolio-level loadings would be a natural next step.
  • The surplus identity suggests a practical underwriting dashboard: for any candidate deployment, compute the four scalar terms and show exactly how far the deployment is from the insurability boundary.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper proposes a mathematical framework for underwriting, pricing, and contract design for agentic-AI deployments. Each deployment is represented by a risk state s_i = (α, β, η, g, v) capturing autonomy category, operational authority, permission profile, governance tier, and dependency concentration. These states are mapped to event probabilities, severities, governance costs, and risk loadings. Insurance contracts are formalized with coverage-incidence matrices, indemnity allocation, deductibles, limits, aggregates, and governance covenants. The insurer's contract-design problem (Problem 3) maximizes risk-adjusted profit subject to participation, profitability, governance, and incentive-compatibility constraints. The paper's central structural claims are: insurability forms a region of the risk-state space (Definition 4), fixed-terms feasibility deteriorates monotonically with exposure (Proposition 5), and a governance threshold certifies insurability (Proposition 6). The key closed form is Eq. (20), where the marketable surplus Δ = J_0 − K(g) − Σ q^e x^e − ρ determines whether a premium interval is nonempty. A healthcare case study and a discrete-event simulation illustrate the framework.

Significance. If the structural results were established for the full Problem 3, the framework would be a useful formalization of an emerging risk class. The paper gives a transparent separation of coverage availability (Γ) from loss allocation (Λ), formalizes governance covenants as contractible obligations, and derives an interpretable marketability condition. The algebraic steps in Eq. (20), Proposition 5, and Proposition 6 are internally correct under the stated monotonicity assumptions. The case study and simulator are helpful for communicating the intended workflow. The main limitation is that the core insurability region and certification threshold are derived without the governance incentive-compatibility constraint that the paper itself states as central; until that gap is closed, the theoretical contribution is conditional. The numerical values are explicitly illustrative rather than empirically calibrated, which is appropriate at this stage but limits the strength of the quantitative conclusions.

major comments (2)
  1. [§5.5, Definition 4, Eq. (20); Prop. 6] Definition 4 defines the insurability region using only the premium interval from constraints (18b)–(18c). The 'if and only if' in Definition 4 is not warranted because Problem 3 also requires (18e): the insured must choose the required governance tier. Eq. (20) cancels the indemnity terms, so Δ(s) is independent of the contract's coverage and of the deviation cost J_gov(˜g; C_i) in Eq. (17). Even if Φ(g) is nonincreasing as assumed in Prop. 6, J_gov need not be minimized at the tier used to compute Δ; the risk loading enters through the premium and the indemnity depends on severity, so the cancellation that makes Δ clean does not apply to the incentive constraint. Hence a state can satisfy Δ ≥ s_0 with no admissible incentive-compatible contract. Please either relabel S_ins as a 'premium-feasible' region or extend the analysis to the full constraint set.
  2. [§7, Problem (23), Table 11] The case-study optimization (23a)–(23g) omits constraint (18e). The finite-menu search over governance tiers g(3) and g(4) and the premium-interval computation never evaluate J_gov(g; C_i) for g(1), ..., g(4); thus the selected contract in Table 11 is not certified as a feasible solution to Problem 3. This is not merely numerical: the text claims the framework jointly handles governance incentives (Section 5.3), but the case study solves a relaxed problem. In addition, Problem 3 lists T_i and D_i^r as scalar decision variables while (18e) refers to schedules τ_i(·) and d_i^r(·); those schedules are not declared as decision variables. Please clarify the formal role of the schedules and either solve the full problem or explicitly state that the case study solves a relaxation without the incentive-compatibility constraint.
minor comments (4)
  1. [§10, Conclusion] The conclusion states that the 'numerical comparisons confirm the structural results.' Since Table 12 is generated from the same monotone functional forms (Eqs. (6)–(7)) that Propositions 5–6 assume, these comparisons illustrate rather than independently confirm the theorems. Suggest rewording to 'instantiate' or 'are consistent with.'
  2. [§4.2, Proposition 2] Proposition 2 is a direct consequence of the maps being functions of s_i; it may be more appropriately framed as a definition of state sufficiency than as a substantive proposition. As it stands, it does not address whether s_i captures all insurance-relevant variation (e.g., model quality, management behavior), which is a separate assumption.
  3. [§7, Table 12] The 'Stronger governance' row changes the required tier from g(3) to g(4), which also changes the governance credit c(g(4)) in the risk-loading rule and the governance cost K_i(g(4)). The comparison is therefore not a pure governance-mitigation effect relative to the baseline; it bundles the loading credit with the loss reduction. This is acceptable for an illustration, but the text should note the loaded components.
  4. [§9.1, Table 15] It would clarify the simulation to explain why 25 automated detections yield only 16 claim notices and 7 denials (i.e., which detections do not become notices) and why human-reviewed claims is zero despite the escalation categories in Figure 8. Adding these details would make the workflow example easier to audit.

Circularity Check

1 steps flagged · score 3.0 of 10

No load-bearing circularity in the formal derivation; one self-consistency overclaim in the case study's 'confirmation' of the structural results.

  1. fitted input called prediction [Section 7, Table 12/Figure 7; Section 10 Conclusion]
    "These one-factor movements instantiate the structural results of Section 5.5: every exposure increase shrinks the marketable surplus toward infeasibility, as in Proposition 5, while the governance upgrade expands it, as in Proposition 6."

    The sensitivity scenarios in Table 12 are computed with the same benchmark exposure-monotone maps Q^e(s_i) (Eq. 6) and X^e(s_i) (Eq. 7) and the same risk-loading rule (including a governance credit for g^(4)) that constitute the hypotheses of Propositions 5 and 6. The numerical 'confirm' in the Conclusion is therefore generated by the assumptions it is said to confirm; it is a self-consistency demonstration, not independent evidence. The formal propositions remain conditional theorems and are not circular, but the case study cannot validate them.

full rationale

The formal derivation chain is self-contained. Equation (20) is algebraically obtained from the participation cap (18b) and the risk-loaded premium floor (18c), with indemnity terms cancelling; Definition 4 and Propositions 5–6 state consequences of the explicitly assumed exposure-monotone and net-risk-reducing primitives. No parameter is fitted to a held-out prediction and no theorem depends on a self-citation: [37] and [38] appear only as related work (Sections 2 and 3), and the proofs use only the paper's own definitions. The one genuine circularity-adjacent step is rhetorical: the healthcare sensitivity analysis is produced from the same functional forms and loading rule assumed in Propositions 5–6, so the Conclusion's phrase 'numerical comparisons confirm the structural results' overstates; it is a self-consistency check. Separately, and not as circularity: Definition 4 and Proposition 6 characterize insurability using only (18b)–(18c), dropping the incentive-compatibility constraint (18e) of Problem 3, so the 'insurability region' is weaker than the full feasibility problem; this is a scope/correctness caveat rather than a circular reduction.

Assumptions & free parameters 9 free parameters · 6 assumptions · 1 invented entities

The framework's quantitative outputs rest on a large set of hand-selected parameters (probability/severity map coefficients, risk weights, loading constants, governance costs, baseline cost, and simulation thresholds). The paper is transparent that most values are illustrative, but the central structural claims inherit the assumed monotonicity of those maps. No free parameters are fitted to real claims data.

free parameters (9)
  • Event-probability map coefficients (a_0^e, a_β^e, a_η^e, a_R^e, qbar^e, qmax^e) in Eq. (6) = Not specified globally; case-study values in Table 9
    The logistic map is a 'benchmark specification' chosen by the author; no estimation from claims data.
  • Severity map coefficients (b_0^e, b_β^e, b_η^e, b_R^e, ξ_α^e, χ^e) in Eq. (7) = Not specified globally; case-study values via Table 9
    Multiplicative severity form is assumed; parameters are said to be calibrated from scorecards/expert elicitation, with no data shown.
  • Permission risk weights ω_j^η (Table 4) = 1, 2, 4, 8, 15
    Illustrative relative weights chosen to rank permission classes; not derived from loss data.
  • Risk-loading rule constants (9000, 0.05, 0.002, 2500, c(g(3))=0, c(g(4))=1500) = Given in Section 7
    Hand-specified loading rule for the case study; no portfolio validation.
  • Governance cost schedule K_i(g) = $30,000 at g(3) in case study
    Assumed annual cost for the governance tier; no market data cited.
  • Uninsured baseline J_i^0 = $100,000 in case study
    Counterfactual assumed to compute the participation constraint; not measured.
  • Case-study event probabilities q_i^e and severities x_i^e (Table 9) = See Table 9
    Representative scenario values, explicitly labeled as not empirical estimates.
  • Case-study menu constraints (75% minimum indemnity ratio, $5,000 minimum surplus, etc.) = As stated in Section 7
    Imposed by the author to avoid thin coverage; choices affect the feasible contract set.
  • Discrete-event simulation thresholds (detection score, evidence score, ambiguity score, trigger thresholds) = Not fully specified
    The simulation is described qualitatively; exact thresholds are not listed, limiting reproducibility.
assumptions (6)
  • domain assumption Exposure-monotone pricing primitives (Prop. 5): q^e, x^e, and ϱ are nondecreasing under the exposure order ⪯_E.
    The structural monotonicity result is conditional on this assumption; it is not derived from data. Section 5.5 states it explicitly.
  • domain assumption Stronger governance is net risk-reducing (Prop. 6): Φ(g) = K(g) + Σ q^e x^e + ϱ is nonincreasing in governance tier.
    The governance certification threshold exists only if the cost of extra governance does not outweigh its risk reduction. Section 5.5 assumes this over the 'relevant range.'
  • domain assumption Risk-state sufficiency (Prop. 2): all insurance-relevant variation is captured by s_i = (α_i, β_i, η_i, g_i, v_i) under the fixed maps.
    Unobserved factors such as model quality, human behavior, or organizational culture are not in the state vector. Section 4.2 formalizes sufficiency by fiat.
  • ad hoc to paper Benchmark functional forms (Eqs. 6–7): logistic probability and multiplicative severity specifications.
    The paper calls these 'transparent benchmark specifications, not the only admissible functional forms'; they are posited, not derived from first principles or data.
  • standard math Standard expected-utility participation and incentive-compatibility (Rothschild–Stiglitz, Ehrlich–Becker).
    The insured's participation constraint and the governance incentive-compatibility condition rely on standard rational-choice assumptions from insurance economics.
  • domain assumption The event taxonomy E = {e_H, e_P, e_F, e_D, e_O, e_C} covers the material loss space for agentic-AI insurance.
    Unlisted failure modes are not priced; the taxonomy is a modeling choice, not empirically demonstrated to be exhaustive.
invented entities (1)
  • AI-insurability certificate (Definition 7)
    purpose: Formalizes the governance threshold g_cert(s_{-g}) as a contractible admission condition; proposed as a regulatory/financial-responsibility instrument.
    The certificate has no falsifiable handle outside the framework; its existence and meaning depend entirely on the assumed monotone maps and the chosen threshold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AI-Native Insurance for Agentic AI: Pricing, Underwriting, and End-to-End Automation." pith.science (2026). https://pith.science/paper/T6I46NEJ

@misc{pith2026260713230,
  author       = {Pith},
  title        = {Pith review of: AI-Native Insurance for Agentic AI: Pricing, Underwriting, and End-to-End Automation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T6I46NEJ}},
  note         = {Machine review of arXiv:2607.13230}
}
read the original abstract

Agentic AI introduces new insurance challenges because autonomous AI systems can make decisions, invoke tools, modify external environments, and interact with third-party services. This paper develops an AI-native mathematical framework for underwriting, pricing, and contract design for agentic AI deployments. A deployment is represented by a risk state that captures autonomy level, operational authority, permission exposure, governance maturity, and dependency concentration. The framework maps the risk state to event probabilities, loss severities, governance costs, premiums, deductibles, coverage allocation, and policy covenants, and formulates an optimization problem for insurance contract design under participation, profitability, and incentive compatibility constraints. The paper establishes structural properties of insurability, including characterization of an insurability region, monotone deterioration of feasibility with increasing exposure, and governance certification thresholds. Insurance is further interpreted as both an operational cost and a regulatory mechanism for AI deployment. A healthcare case study illustrates contract optimization, sensitivity analysis, and automated claims processing for agentic AI systems.

Figures

Figures reproduced from arXiv: 2607.13230 by the authors.

Figure 1
Figure 1. Illustrative bar-plot examples of risk mappings evaluated at discrete governance tiers. [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Illustrative bar-plot examples of financial and contract-schedule mappings evaluated at [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Illustrative layer-level indemnity schedule [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Illustrative total risk-loading levels for four underwriting scenarios. The values are [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: Illustrative governance-deviation cost J gov i (g; Ci) evaluated over discrete governance tiers. Each stacked bar decomposes total expected cost into the governance-sensitive premium, governance-control cost, and retained expected loss. In this example, the premium and…
Figure 6
Figure 6. Figure 6: Computational solution of the healthcare case study. The left panel plots all feasible [PITH_FULL_IMAGE:figures/full_fig_p037_6.png]
Figure 7
Figure 7. Figure 7: Sensitivity of the optimized healthcare contract to selected underwriting variables. The [PITH_FULL_IMAGE:figures/full_fig_p038_7.png]
Figure 8
Figure 8. Figure 8: Automated insurance workflow for insured agentic-AI systems. The workflow be [PITH_FULL_IMAGE:figures/full_fig_p043_8.png]
Figure 9
Figure 9. Figure 9: Discrete-event simulation of online automated claim execution for a small portfolio of [PITH_FULL_IMAGE:figures/full_fig_p046_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 9 linked inside Pith

  1. [1]

    Amodei, C

    D. Amodei, C. Olah, J. Steinhardt, P. Christiano, J. Schulman, and D. Mane. Concrete problems in ai safety.arXiv preprint arXiv:1606.06565, 2016

    arXiv 2016

  2. [2]

    Armilla launches affirmative ai liability insurance with lloyd’s underwriter chaucer,

    Armilla AI. Armilla launches affirmative ai liability insurance with lloyd’s underwriter chaucer,

  3. [3]

    AXA XL unveils new cyber insurance extending coverage to help businesses manage emerging GenAI risks, 2024

    AXA XL. AXA XL unveils new cyber insurance extending coverage to help businesses manage emerging GenAI risks, 2024. Public announcement

    2024

  4. [4]

    E. M. Bender, T. Gebru, A. McMillan-Major, and S. Shmitchell. On the dangers of stochastic parrots: Can language models be too big? InProceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pages 610–623, 2021

    2021

  5. [5]

    Bengio, G

    Y. Bengio, G. Hinton, A. Yao, D. Song, P. Abbeel, T. Darrell, Y. N. Harari, Y.-Q. Zhang, L. Xue, and S. Shalev-Shwartz. Managing extreme ai risks amid rapid progress.Science, 384(6698):842–845, 2024

    2024

  6. [6]

    Biener, M

    C. Biener, M. Eling, and J. H. Wirfs. Insurability of cyber risk: An empirical analysis.The Geneva Papers on Risk and Insurance - Issues and Practice, 40(1):131–158, 2015

    2015

  7. [7]

    Bohme and G

    R. Bohme and G. Schwartz. Modeling cyber-insurance: Towards a unifying framework. In Workshop on the Economics of Information Security, 2010

    2010

  8. [8]

    Bolot and M

    J. Bolot and M. Lelarge. Cyber insurance as an incentive for internet security. In M. E. Johnson, editor,Managing Information Risk and the Economics of Security, pages 269–290. Springer, New York, 2009

    2009

Show all 38 references
  1. [9]

    Bommasani, D

    R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, et al. On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258, 2021

    2021 arXiv

  2. [10]

    Calabresi.The Costs of Accidents: A Legal and Economic Analysis

    G. Calabresi.The Costs of Accidents: A Legal and Economic Analysis. Yale University Press, 1970

    1970

  3. [11]

    Carlini and D

    N. Carlini and D. Wagner. Towards evaluating the robustness of neural networks. In2017 IEEE Symposium on Security and Privacy, pages 39–57, 2017

    2017

  4. [12]

    Chaucer and armilla ai launch vanguard ai coordinated insurance structure, 2025

    Chaucer Group and Armilla AI. Chaucer and armilla ai launch vanguard ai coordinated insurance structure, 2025. Public announcement

    2025

  5. [13]

    J. Chen, Q. Zhu, and T. Ba¸ sar. Dynamic contract design for systemic cyber risk management of interdependent enterprise networks.Dynamic Games and Applications, 11(2):294–325, 2021

    2021

  6. [14]

    Ehrlich and G

    I. Ehrlich and G. S. Becker. Market insurance, self-insurance, and self-protection.Journal of Political Economy, 80(4):623–648, 1972

    1972

  7. [15]

    Methodological principles of insur- ance stress testing: Cyber component, 2024

    European Insurance and Occupational Pensions Authority. Methodological principles of insur- ance stress testing: Cyber component, 2024. Online report

    2024

  8. [16]

    I. J. Goodfellow, J. Shlens, and C. Szegedy. Explaining and harnessing adversarial examples. InInternational Conference on Learning Representations, 2015

    2015

  9. [17]

    L. A. Gordon and M. P. Loeb. The economics of information security investment.ACM Transactions on Information and System Security, 5(4):438–457, 2002

    2002

  10. [18]

    Greshake, S

    K. Greshake, S. Abdelnabi, S. Mishra, C. Endres, T. Holz, and M. Fritz. Not what you’ve signed up for: Compromising real-world llm-integrated applications with indirect prompt injection. InProceedings of the 16th ACM Workshop on Artificial Intelligence and Security, pages 79–90, 2023

    2023

  11. [19]

    Humayed, J

    A. Humayed, J. Lin, F. Li, and B. Luo. Cyber-physical systems security–a survey.IEEE Internet of Things Journal, 4(6):1802–1831, 2017

    2017

  12. [20]

    Liu and Q

    S. Liu and Q. Zhu. Mitigating moral hazard in cyber insurance using risk preference design. arXiv preprint arXiv:2203.12001, 2022

    2022 arXiv

  13. [21]

    Liu and Q

    S. Liu and Q. Zhu. Cyber insurance for cyber resilience.arXiv preprint arXiv:2312.02921, 2023

    2023 arXiv

  14. [22]

    Naghizadeh and M

    P. Naghizadeh and M. Liu. A tale of two mechanisms: Incentivizing investments in security games.arXiv preprint arXiv:1503.07377, 2015

    2015 arXiv

  15. [23]

    Cybersecurity insurance report, 2025

    National Association of Insurance Commissioners. Cybersecurity insurance report, 2025. On- line report

    2025

  16. [24]

    Artificial intelligence risk management frame- work (ai rmf 1.0)

    National Institute of Standards and Technology. Artificial intelligence risk management frame- work (ai rmf 1.0). Technical Report NIST AI 100-1, National Institute of Standards and Technology, 2023

    2023

  17. [25]

    OWASP Top 10 for Large Language Model Applications.https:// owasp.org/www-project-top-10-for-large-language-model-applications/, 2025

    OWASP Foundation. OWASP Top 10 for Large Language Model Applications.https:// owasp.org/www-project-top-10-for-large-language-model-applications/, 2025

    2025

  18. [26]

    R. Pal, Z. Huang, X. Yin, S. Lototsky, S. De, S. Tarkoma, M. Liu, J. Crowcroft, and N. Sastry. Aggregate cyber-risk management in the IoT age: Cautionary statistics for (re)insurers and likes.arXiv preprint arXiv:2105.01792, 2021

    2021 arXiv

  19. [27]

    Perez and I

    F. Perez and I. Ribeiro. Ignore previous prompt: Attack techniques for language models.arXiv preprint arXiv:2211.09527, 2022

    2022 arXiv

  20. [28]

    A. C. Pigou.The Economics of Welfare. Macmillan, 1920

    1920

  21. [29]

    Quinonero-Candela, M

    J. Quinonero-Candela, M. Sugiyama, A. Schwaighofer, and N. D. Lawrence, editors.Dataset Shift in Machine Learning. MIT Press, Cambridge, MA, 2009

    2009

  22. [30]

    A. Raviv. The design of an optimal insurance policy.The American Economic Review, 69(1):84–96, 1979

    1979

  23. [31]

    Romanosky, L

    S. Romanosky, L. Ablon, A. Kuehn, and T. Jones. Content analysis of cyber insurance policies: How do carriers price cyber risk?Journal of Cybersecurity, 5(1):tyz002, 2019

    2019

  24. [32]

    Rothschild and J

    M. Rothschild and J. Stiglitz. Equilibrium in competitive insurance markets: An essay on the economics of imperfect information.The Quarterly Journal of Economics, 90(4):629–649, 1976

    1976

  25. [33]

    S. Shavell. A model of the optimal use of liability and safety regulation.The RAND Journal of Economics, 15(2):271–280, 1984

    1984

  26. [34]

    Advancing accumulation risk management in cyber insurance, 2024

    The Geneva Association. Advancing accumulation risk management in cyber insurance, 2024. Online report

    2024

  27. [35]

    Zhang and Q

    R. Zhang and Q. Zhu. Attack-aware cyber insurance of interdependent computer networks. Technical Report 16-18, NET Institute, 2016

    2016

  28. [36]

    Zhang and Q

    R. Zhang and Q. Zhu. Optimal cyber-insurance contract design for dynamic risk management and mitigation.IEEE Transactions on Computational Social Systems, 9(4):1087–1100, 2021

    2021

  29. [37]

    Q. Zhu. Insurance of agentic ai, 2026. arXiv preprint arXiv:2606.05449, 3 June 2026

    2026 arXiv

  30. [38]

    Q. Zhu. The internet of agentic ai: Communication, coordination, and collective intelligence at scale, 2026. arXiv preprint arXiv:2606.12835, 11 June 2026

    2026 arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.