pith. sign in

arxiv: 2604.28113 · v2 · submitted 2026-04-30 · 💻 cs.CY · cs.AR· cs.SE

I hope we don't do to trust what advertising has done to love

Jade Alglave This is my paper

Pith reviewed 2026-05-14 22:01 UTC · model grok-4.3

classification 💻 cs.CY cs.ARcs.SE
keywords trust in AItrust pillarstrust vectorsagentic systemsAI ethicscomputing and societyconceptual decomposition
0
0 comments X

The pith

Trust in AI should be broken into measurable pillars and vectors from agentic interfaces rather than left vague like advertising's use of love.

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

The paper warns that talk of trust in AI, especially agentic systems, could lose all substance the way advertising emptied out the word love through casual overuse. It proposes defining a set of trust pillars to turn the topic into something actionable and measurable that computing experts and civil society can actually discuss and apply. The author also sees agentic systems as potentially helpful because their explicit interfaces could be treated as trust vectors that make verification concrete. This matters if true because without such structure, decisions about deploying AI will stay based on empty assurances rather than workable checks. The goal is to start a shared conversation that keeps trust discussions grounded and useful across fields.

Core claim

The author suggests defining a number of trust pillars to allow trust in AI to be discussed in actionable and measurable ways, starting a conversation across computing and civil society. Agentic systems may be a blessing because their explicit interfaces can be turned into trust vectors that operationalize trust instead of leaving it abstract.

What carries the argument

Trust pillars as a proposed decomposition of trust into actionable components, and trust vectors as the explicit interfaces of agentic systems that can serve to build and verify trust.

If this is right

  • Trust discussions in AI shift from abstract to concrete and measurable terms.
  • Agentic systems supply ready-made interfaces that can function as trust vectors.
  • Conversations about AI trust expand to include both technical fields and civil society.
  • AI development can prioritize explicit interface design to support trust verification.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The pillar approach could extend to other hard-to-pin-down ideas such as fairness or safety in technology.
  • Standardized trust metrics might emerge if the pillars receive concrete examples and testing.
  • Agentic systems would need deliberate interface choices to maximize their value as trust vectors.
  • Without quick examples of pillars the suggestion risks staying at the level of a call for discussion.

Load-bearing premise

That trust in AI can be usefully decomposed into actionable and measurable pillars and vectors without first providing concrete definitions, examples, or validation methods.

What would settle it

A pilot project that defines specific trust pillars, applies them to an agentic AI system, and then measures whether resulting discussions or trust assessments become more concrete and verifiable than before.

Figures

Figures reproduced from arXiv: 2604.28113 by Jade Alglave.

Figure 1
Figure 1. Figure 1: Basic agentic architecture and Examples of trust-related questions view at source ↗
Figure 2
Figure 2. Figure 2: Trust pillars into unrelated data collection or submission authority; Legibility: the renewal state, e.g. requirements and next steps, are intelligible to users; Contestability: there exists a practically feasible route to challenge rejection or misprocessing; Redress: there exists a practically feasible path to seek correction to processing error, or submission failure; Survivability: under uncertainty or… view at source ↗
Figure 3
Figure 3. Figure 3: A few suggestions I found inspiration for these whilst reflecting on what trust might mean for users like my grandma in the context of agentic AI, imagining what mechanisms might help meet the trust pillars. To let us imagine further, here are a few papers that have resonated with me: • Adabara et al [1] provide a “cross-layer review of agentic AI, encompassing architectural paradigms, threat tax￾onomies, … view at source ↗
read the original abstract

Advertising uses love to sell stuff, like nylons. It also uses the word "love" in trivialising ways -- do you "love" your oven? When I hear about trust in the context of AI, especially agentic, I hope we don't do to trust what advertising has done to love. But what is trust? Can we discuss it in actionable and measurable ways in the context of AI? Thus I suggest a number of "trust pillars", hoping to start a communal conversation, across computing and beyond, to civil society. I also suggest that agentic systems may be a blessing in disguise, as we may be able to turn their explicit interfaces into "trust vectors".

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

0 major / 3 minor

Summary. The manuscript is a short position paper warning that discussions of trust in AI risk trivializing the concept in the same way advertising has trivialized 'love'. It proposes a set of 'trust pillars' to enable actionable and measurable conversations about trust in AI systems and suggests that the explicit interfaces of agentic AI systems could function as 'trust vectors' to spark broader dialogue across computing and civil society.

Significance. If the proposed pillars and vectors successfully initiate sustained, cross-disciplinary discussion, the paper could help steer AI research and policy toward more substantive treatments of trust. Its contribution is primarily invitational rather than evidentiary, which is appropriate for a position piece in the computers-and-society literature; its influence will depend on whether subsequent work supplies the missing definitions, metrics, and examples.

minor comments (3)
  1. The manuscript refers to 'a number of trust pillars' without enumerating or briefly characterizing them; adding even a short list or one-sentence glosses would give readers concrete entry points for the intended conversation.
  2. The title is informal and allusive; a subtitle or parenthetical clarification would help readers quickly grasp the central analogy and proposal.
  3. No references to prior work on trust in AI (e.g., in human-computer interaction or AI ethics) are provided; a brief contextualizing paragraph would strengthen the positioning of the new suggestions.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their review and positive recommendation of minor revision. We appreciate the recognition that the contribution is invitational and appropriate for a position paper in the computers-and-society literature. No specific major comments appear in the report, so there are no individual points requiring direct response or revision at this stage. We will incorporate any editorial suggestions during the minor revision process.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript is a short position paper that proposes 'trust pillars' and 'trust vectors' explicitly as an invitation to start a communal conversation rather than as derived or validated constructs. No equations, fitted parameters, self-citations, or definitional loops appear; trust is not defined in terms of the proposed pillars inside the argument, and the text frames its contribution as open-ended discussion without claiming predictive or first-principles derivations that reduce to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The contribution rests on two newly introduced concepts and the assumption that trust admits decomposition into measurable pillars; no free parameters or external benchmarks are used.

axioms (1)
  • domain assumption Trust in AI can be discussed in actionable and measurable ways
    Stated as the motivation for proposing pillars without supporting evidence or prior validation.
invented entities (2)
  • trust pillars no independent evidence
    purpose: To structure discussions of trust in AI into measurable components
    New term introduced to operationalize trust; no independent evidence or prior definition supplied.
  • trust vectors no independent evidence
    purpose: To use explicit interfaces of agentic systems as measurable indicators of trust
    New concept tied to agentic AI; presented without empirical grounding or falsifiable test.

pith-pipeline@v0.9.0 · 5405 in / 1257 out tokens · 70874 ms · 2026-05-14T22:01:18.929643+00:00 · methodology

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages

  1. [1]

    Trustworthy agentic AI systems: a cross-layer review of architectures, threat models, and governance strategies for real-world deployment

    Ibrahim Adabara, Bashir Olaniyi Sadiq, Aliyu Nuhu Shuaibu, Yale Ibrahim Danjuma, and Venkateswarlu Maninti. Trustworthy agentic AI systems: a cross-layer review of architectures, threat models, and governance strategies for real-world deployment. 2025.https://f1000research.com/articles/14-905/v1

    work page 2025

  2. [2]

    Confidential Compute Architecture

    Arm. Confidential Compute Architecture. 2026.https://www.arm.com/architecture/security-features/ arm-confidential-compute-architecture

    work page 2026

  3. [3]

    Small Language Models are the Future of Agentic AI

    Peter Belcak, Greg Heinrich, Shizhe Diao, Yonggan Fu, Xin Dong, Saurav Muralidharan, Yingyan Celine Lin, and Pavlo Molchanov. Small Language Models are the Future of Agentic AI. 2025.https://arxiv.org/abs/2506. 02153

    work page 2025

  4. [4]

    Toward adaptive categories: dimensional governance for agentic AI

    Zeynep Engin and David Hand. Toward adaptive categories: dimensional governance for agentic AI. 2025.https: //arxiv.org/abs/2505.11579

    work page arXiv 2025

  5. [5]

    C2pa Specifications

    Coalition for Content Provenance and Authenticity. C2pa Specifications. 2026.https://spec.c2pa.org/ specifications/specifications/2.2/index.html

    work page 2026

  6. [6]

    AI Cybersecurity After Mythos: The Jagged Frontier

    Stanislav Fort. AI Cybersecurity After Mythos: The Jagged Frontier. 2026.https://aisle.com/blog/ ai-cybersecurity-after-mythos-the-jagged-frontier

    work page 2026

  7. [7]

    Linda Childers Hon and James E. Grunig. Guidelines for measuring relationships in public relations. 1999.https: //instituteforpr.org/wp-content/uploads/Guidelines_Measuring_Relationships.pdf

    work page 1999

  8. [8]

    On Working With Wizards

    Ethan Mollick. On Working With Wizards. 2026.https://www.oneusefulthing.org/p/ on-working-with-wizards

    work page 2026

  9. [9]

    Trust, Trustworthiness And Trans- parency

    Onora O’Neill and James Bardrick. Trust, Trustworthiness And Trans- parency. 2017.https://www.thebritishacademy.ac.uk/documents/2563/ Future-of-the-corporation-Trust-trustworthiness-transparency.pdf

    work page 2017

  10. [10]

    Engineering trustworthy secure systems NIST SP 800-160 vol- ume 1 revision 1

    Ron Ross, Mark Winstead, and Michael McEvilley. Engineering trustworthy secure systems NIST SP 800-160 vol- ume 1 revision 1. 2022.https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-160v1r1. pdf. 5

    work page 2022

  11. [11]

    Agents of chaos

    Natalie Shapira, Chris Wendler, Avery Yen, Gabriele Sarti, Koyena Pal, Olivia Floody, Adam Belfki, Alex Lof- tus, Aditya Ratan Jannali, Nikhil Prakash, Jasmine Cui, Giordano Rogers, Jannik Brinkmann, Can Rager, Amir Zur, Michael Ripa, Aruna Sankaranarayanan, David Atkinson, Rohit Gandikota, Jaden Fiotto-Kaufman, EunJeong Hwang, Hadas Orgad, P Sam Sahil, N...

    work page 2026

  12. [12]

    Robinson

    Yonadav Shavit, Sandhini Agarwal, Miles Brundage, Steven Adler, Cullen O’Keefe, Rosie Campbell, Teddy Lee, Pamela Mishkin, Tyna Eloundou, Alan Hickey, Katarina Slama, Lama Ahmad, Paul McMillan, Alex Beutel, Alexan- dre Passos, and David G. Robinson. Practices for governing agentic AI systems. 2023.https://openai.com/ index/practices-for-governing-agentic-...

    work page 2023

  13. [13]

    Directions of technical innovation for regulatable AI systems

    Xudong Shen, Hannah Brown, Jiashu Tao, Martin Strobel, Yao Tong, Akshay Narayan, Harold Soh, and Finale Doshi-Velez. Directions of technical innovation for regulatable AI systems. 2024.https://cacm.acm.org/ research/directions-of-technical-innovation-for-regulatable-ai-systems/

    work page 2024

  14. [14]

    Large language models can solve some instances of the halting problem

    Oren Sultan, Jordi Armengol-Estape, Pascal Kesseli, Julien Vanegue, Dafna Shahaf, Yossi Adi, and Peter O’Hearn. Large language models can solve some instances of the halting problem. 2026.https://arxiv.org/abs/2601. 18987

    work page 2026

  15. [15]

    The Bitter Lesson

    Richard Sutton. The Bitter Lesson. 2019.http://www.incompleteideas.net/IncIdeas/BitterLesson.html

    work page 2019

  16. [16]

    Trust in AI: global study on sentiments and at- titudes towards AI

    University of Melbourne and KPMG International. Trust in AI: global study on sentiments and at- titudes towards AI. 2025.https://mbs.edu/-/media/PDF/Research/Trust_in_AI_Report.pdf?rev= 0ee82285b2b0439bba524dbddc58214a

    work page 2025

  17. [17]

    This is not the AI we were promised

    Michael Wooldridge. This is not the AI we were promised. 2026.https://royalsociety.org/ science-events-and-lectures/2026/02/faraday-prize-lecture/. 6

    work page 2026