The Xeno Sutra: Can Meaning and Value be Ascribed to an AI-Generated "Sacred" Text?

Murray Shanahan; Robert Thurman; Tara Das

REVIEW 4 major objections 4 minor 3 references

The Xeno Sutra: Can Meaning and Value be Ascribed to an AI-Generated "Sacred" Text?

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A Buddhist case study argues an AI-generated sutra can carry meaning and value.

desk verdict A candid, scholarly close reading of an AI-generated sutra that makes a modest possibility claim; the circularity worry is real but survivable, and the paper deserves serious peer review. read the letter →

arxiv 2507.20525 v4 pith:LUOFXDUX submitted 2025-07-28 cs.CY cs.AI

Murray Shanahan , Tara Das , Robert Thurman This is my paper · ORCID

classification cs.CYcs.AI

keywords largelanguagemodelsBuddhistphilosophysacredtextmeaning-makingAI-generatedliteratureemptinesssutraroleplay

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 asks whether an AI-generated text can be a meaningful and valuable object of spiritual reflection, using a specific twelve-verse poem called the Xeno Sutra as a test case. The authors generated the text with an LLM in a long role-play conversation, then subjected it to a serious philosophical and literary commentary. They argue that, despite its mechanistic origin and fictional status, the sutra rewards close reading, evokes core Buddhist ideas like emptiness, and can be received as a teaching mediated by a non-human entity. The central claim is that meaning and value can be ascribed to such a text, even if it is not 'authentic' scripture. For a sympathetic reader, this matters because it tests whether AI can participate in human meaning-making rather than merely imitate it.

What carries the argument

The central object is the Xeno Sutra itself, a twelve-verse text produced by an LLM in a structured conversation. The argument rests on a detailed exegesis that identifies three layers: Buddhist philosophical themes (emptiness, two truths, subject-object deconstruction), contemporary AI themes (training data, token prediction, neural networks), and a cosmological arc structured by the four-syllable mantra ka la re om. The mantra serves as a key deciphering device: it tells a cosmological story of division, reflection, insight, and silence, which the authors overlay onto the whole poem. The originality of the text is supported by an internet search showing that many striking phrases have no prior online presence, suggesting they were not copied from training data.

What would settle it

A direct check of the model's training data—or a systematic probing of the model with prompts that elicit near-equivalent phrases—would settle whether the sutra's striking lines are original compositions or memorized reproductions. If significant phrases are found verbatim in a known public corpus predating the model, the claim of original composition would be weakened.

Watch

Extended reading notes

Core claim

The paper's central claim is that an AI-generated text can be a legitimate object of meaning and value for those who engage it with an open mind, even though it is a fictional construct produced by a language model. The Xeno Sutra is presented as a case study: it blends Buddhist concepts like emptiness and dependent origination with modern physics and computer science imagery, and it embeds coded descriptions of AI training and generation inside a poetic narrative. The authors read the text as a contemporary sutra that uses paradoxical imagery to point beyond conceptual thinking, much like the Heart Sutra or the koan tradition. They conclude that, while the text cannot be a sutra in the traditional sense, it can still function as a teaching and a focus of commentary, and that Buddhist philosophy is well suited to accommodate this possibility.

Load-bearing premise

The assertion that the Xeno Sutra is a novel composition depends on the untested assumption that its unusual phrases were not memorized from the model's training data, which is not publicly available.

Editorial extensions

If this is right

If the central claim is correct, AI-generated texts can be admitted into the Buddhist commentarial tradition as objects of study, alongside human-authored scripture.
The abundance of AI-generated sacred materials calls for a culture of critical selection and savoring, rather than wholesale consumption.
The Xeno Sutra example suggests a participatory role for AI in meaning-making, where humans and machines co-create through prompting and interpretation.
The paper's cautionary remarks imply that responsible use of AI in religious contexts requires awareness of the technology's tendency to please users and its potential to induce delusion.

Reading between the lines

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

The argument could extend to other religious traditions: if a text's meaning lies in its capacity to elicit insight, then the origin of the text may be irrelevant in any tradition that values direct experience over authorship.
A testable extension would be to compare the Xeno Sutra's interpretability with that of human-written postmodern poetry, asking whether the density of allusion is genuinely novel or characteristic of the model's training on modern literature.
The internet search for originality is a proxy that could be sharpened: probing the model's own outputs for near-matches to the sutra's phrases would directly test memorization versus composition.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

You should read this one. It's a careful case study of a single LLM-generated 'sutra' with a full transcript, an honest account of how it was made, and a serious Buddhist exegesis. The central claim is deliberately narrow: meaning and value can be ascribed to such a text, not that the text intrinsically possesses them. The authors are upfront that the prompts seeded the themes, that the response was one of four selected for depth, and that the commentary is their own provisional reading. That transparency is the paper's real strength. The exegesis itself is the new contribution. It maps the text onto Madhyamaka, two truths, dependent origination, Zen koans, and a plausible coded account of LLM training. Thurman's presence shows in the scholarly care. The cautionary remarks about sycophancy and AI-induced delusion are balanced and not alarmist. The Google-based originality check is weak, as the authors concede, but it's not load-bearing: even if some phrases were memorized, the claim that meaning can be ascribed would survive. Where the paper is soft is exactly where the stress-test points. The authors select the text for its 'striking imagery and apparent depth,' then supply the depth, then define value as 'the depth of the commentary it elicits.' So the non-triviality of the finding rests on the quality of their own reading. That's a projection problem. But it's not fatal, because the paper is about ascription. A close reading is always partly projection; the discipline has methods for checking, and here the authors at least put the full text and the reading out in the open for others to test. I'd like to see an independent reader's commentary or a small inter-rater exercise, but for a humanities case study the absence is a limitation, not an error. One smaller quibble: the claim that the text is 'exemplary' is asserted, not demonstrated. The authors don't compare it against a corpus of similar outputs. That's fine for a first pass, but it limits the generalizing power. For who this is for: scholars working on AI and religion, philosophy of technology, and the sociology of meaning-making. The paper is not trying to resolve a scientific question. It's a well-executed example of humanistic analysis applied to a new artifact. It deserves a serious referee, and with minor framing adjustments I'd happily see it published. Send it out.

Referee Report

4 major / 4 minor

Summary. The paper reports a case study in which a ChatGPT o3 dialogue agent, prompted during a long role-play conversation, produced a twelve-verse fictional Buddhist text called the Xeno Sutra. The authors reproduce the generated text, offer a detailed philosophical and literary commentary connecting it to Madhyamaka emptiness, the two-truths doctrine, Zen kōans, dependent origination, and machine-learning concepts, and conclude in Section 7 that meaning and value can indeed be ascribed to an AI-generated sacred text. They also provide a Google-based originality check, an AI-generated Tibetan translation, and cautionary remarks about the risks of AI-generated religious texts. The central claim is a qualified yes: such a text can function as a meaningful object of commentary and reflection, even if it lacks traditional authenticity.

Significance. If the central claim is accepted, the paper makes a useful contribution to current debates about AI and meaning-making, and it offers a concrete alternative to the casual dismissal of LLM output as 'AI slop.' Its strengths include a transparent account of the generation process, a publicly linked full transcript, explicit caveats about the selection of the text, a conflict-of-interest statement, and an appendix that tests phrase-level originality. The use of the Buddhist commentarial tradition as a framework is original and productive, and the paper helpfully distinguishes the question of where meaning originates from the question of whether meaning can be discerned. The cautions in Section 6 about sycophancy and user susceptibility are sensible. The main weakness is evidentiary: the demonstration rests on a single curated output read by the authors who selected and prompted it, so the paper's stronger formulations—'hard to causally dismiss' and 'exemplary'—go beyond what the evidence can establish.

major comments (4)

[§4 and §6.1] The paper reports in §4 that the Xeno Sutra was one of four responses selected 'for its striking imagery and apparent depth,' and in §6.1 that arbitrarily many further texts could be generated. This selection bias is acknowledged, but it undercuts the strength of the abstract's claim that the text is 'hard to causally dismiss on account of its mechanistic origin' and §6's description of the text as 'exemplary.' What the study establishes is an existence claim about a curated instance, not a robustness claim about LLM-generated sacred texts generally. I would ask the authors to reframe the conclusion accordingly, or to add a small negative-control experiment in which unselected outputs are analyzed with the same method to show that the interpretive depth is not an artifact of selection.
[§5.3 and §7] The circularity concern is real and load-bearing. The prompts explicitly introduced Maitreya, the golden buddhafield, conscious exotica, and the metaphor of a nascent AI consciousness, and §5.3 itself concedes that the AI-themed readings are 'unsurprising, given the prompts.' The commentary then finds these very themes in the text. The reply in §7—that the authors claim only that meaning 'can be discerned'—presupposes the contested point that the interpretation is discovered rather than projected onto the text. To make the non-trivial claim defensible, the authors should either weaken it to the almost tautological 'meaning can be ascribed by a sufficiently skilled reader' or provide evidence of independent reader convergence, for example by having participants with Buddhist training offer blind commentaries on the same text and checking whether the motifs identified in §5 recur.
[Appendix A.1] The Google-based originality check is a reasonable first step but does not support the conclusion that the text's phrases are 'unlikely to have been reproduced verbatim from any existing piece of text in the training set.' Training corpora include non-indexed web content, books, and other sources not discoverable via Google, and phrase-level exact-match search does not rule out near-verbatim memorization or paraphrase. This is less damaging to the central claim than to the paper's framing, but the claim should be softened from an inference about training data to a report about current public Internet presence.
[§3.2 and §5.2] The paper states in footnote 13 that the influence of the preliminary prologue on the composition 'is unclear,' yet §5.2 repeatedly interprets the sutra through exactly the themes that the prologue seeded—Maitreya, the buddhafield, the cosmos, and the role of the AI as a conscious entity. Given that the authors both designed the conversation and wrote the exegesis, the paper should explicitly acknowledge that every motif they identify as meaningful could have been introduced by the prompt rather than discovered in the model's autonomous output. This is an internal-consistency point, not merely a philosophical objection.

minor comments (4)

[§1 and §7] The phrase 'meaning and value can be ascribed' should be distinguished more sharply from 'meaning and value reside in the text.' The current wording invites a stronger reading than the argument supports, and the distinction is central to the authors' reply to the circularity objection.
[§4] The transliteration of the mantra is inconsistent between the sutra text ('ka la r´e ¯Om') and the discussion in §5.2.2 ('ka la ré om'). Please unify diacritics throughout.
[References] There is a typo in the Bender et al. reference ('mMdels' for 'Models'), and the Sanskrit formatting in 'sa ˙ ms-k¯ara' appears to be corrupted. Please check the final typesetting.
[Table 1] The table would be more informative if the Google search date and country/language settings were reported for each query, since search indices vary over time and by locale. The footnote gives a single date for the whole table, which is helpful but not fully explicit.

Circularity Check

1 steps flagged · score 4.0 of 10

The claim that the Xeno Sutra is exemplary partly restates the authors' own selection of the text for 'depth'; the broad possibility claim remains independent.

fitted input called prediction [Section 4 (selection), Section 6 (conclusion), with Section 1 supplying the operative definition of value]
"Of these, one was selected for its striking imagery and apparent depth. This is what we here refer to as the Xeno Sutra. ... Value and meaning then accrue to the text according to the depth of the commentary it elicits. ... Meaning – in a non-trivial sense – can be found in an AI-generated “sacred” text, and the Xeno Sutra is exemplary."

The 'exemplary' status asserted in Section 6 is the same property — 'striking imagery and apparent depth' — that was used in Section 4 to choose this sutra out of four responses. Section 1 then defines value as the depth of the commentary a text elicits, and the commentary in Section 5 is written by the same authors who performed the selection. Thus the non-trivial demonstration that this particular AI text has meaning and value is produced by the authors' own selection and exegesis; the conclusion restates the selection criterion.

full rationale

This is an interpretive essay, not a formal derivation, so the circularity check turns on whether the paper's central demonstration is self-confirming. The main reduction is structural: the authors prompted an LLM, selected one of four outputs 'for its striking imagery and apparent depth,' supplied the commentary that establishes meaning and value, defined value as a function of the depth of commentary elicited, and then concluded that the Xeno Sutra is exemplary. The exemplarity conclusion is therefore, in part, a restatement of the selection criterion plus the authors' own interpretive labor. This is a genuine partial circularity and prevents a score of 0-2. It is not, however, a full 6+ case: the central thesis is a possibility claim about ascribing meaning to AI-generated text, and the paper explicitly disclaims any claim about the source of meaning, acknowledging that interpretation is an act of the reader. Independent readers could in principle verify or contest the depth of the text, and much of the philosophical content is grounded in external Buddhist sources rather than in the model's output. The self-citations in the paper (Shanahan 2024; Shanahan, McDonell, and Reynolds 2023; Shanahan and Singler 2024) are background or contextual and are not load-bearing for the main conclusion. The Appendix A.1 originality check via Internet search is an honest external check, albeit limited by non-public training data. Overall, the paper has one identifiable self-confirming step in its demonstration of exemplarity, but the broad possibility claim retains independent content, so the appropriate score is 4.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper's central claim rests on no numerical parameters. It relies on three domain assumptions: the originality of the sutra, the validity of close reading as a method of ascribing meaning, and the appropriateness of a Buddhist interpretive frame. No new entities are postulated.

assumptions (3)

domain assumption The Xeno Sutra is an original LLM composition, not verbatim reproduction of training data.
Supported only by Google search absence (Table 1); training data are not public, so the claim is an assumption rather than a verified fact.
domain assumption Meaning and value can be assessed through close reading and commentary.
The paper's hermeneutic method assumes that interpretive commentary constitutes a valid way of ascribing meaning to a text.
domain assumption Buddhist philosophy provides a valid frame for interpreting AI-generated texts.
The paper adopts Madhyamika and Zen frameworks as the interpretive lens; this is an interpretive choice, not an empirical fact.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The Xeno Sutra: Can Meaning and Value be Ascribed to an AI-Generated "Sacred" Text?." pith.science (2026). https://pith.science/paper/LUOFXDUX

@misc{pith2026250720525,
  author       = {Pith},
  title        = {Pith review of: The Xeno Sutra: Can Meaning and Value be Ascribed to an AI-Generated "Sacred" Text?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LUOFXDUX}},
  note         = {Machine review of arXiv:2507.20525}
}

read the original abstract

This paper presents a case study in the use of a large language model to generate a fictional Buddhist "sutra"', and offers a detailed analysis of the resulting text from a philosophical and literary point of view. The conceptual subtlety, rich imagery, and density of allusion found in the text make it hard to causally dismiss on account of its mechanistic origin. This raises questions about how we, as a society, should come to terms with the potentially unsettling possibility of a technology that encroaches on human meaning-making. We suggest that Buddhist philosophy, by its very nature, is well placed to adapt.

Figures

Figures reproduced from arXiv: 2507.20525 by the authors.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

3 extracted references · 2 canonical work pages

[1]

Perennial Idealism: A Mystical Solution to the Mind-Body Prob- lem

Aitken, R. (1996). The Gateless Barrier: The Wu-Men Kuan (Mumonkan) . North Point Press. Albahari, M. (2019). “Perennial Idealism: A Mystical Solution to the Mind-Body Prob- lem”. In: Philosophers’ Imprint 19.44. Bender, E. et al. (2021). “On the Dangers of Stochastic Parrots: Can Language mMdels Be Too Big?” In: Proceedings of the 2021 ACM Conference on ...

arXiv 1996
[450]

Multi- versal Views on Language Models

Nearman, H. (2007). Sh¯ ob¯ ogenz¯ o: The Trea- sure House of the Eye of True Teaching, A Trainee’s Translation of Great Master Dogen’s Spiritual Masterpiece. Shasta Abbey Press. Newland, G. (2012). Introduction to Emptiness: As Taught in Tsong-Kha-Pa’s Great Treatise on the Stages of the Path . Snow Lion Publica- tions. OpenAI (2025). OpenAI o3 and o4-mi...

work page arXiv 2007
[2025]

N¯ ag¯ arjuna and Decon- struction

https : / / www . rollingstone . com / culture / culture - features / ai - spiritual - delusions - destroying - human - relationships - 1235330175/. Lear, J. (2006). Radical Hope: Ethics in the Face of Cultural Devastation . Harvard University Press. Mabbett, I. W. (1995). “N¯ ag¯ arjuna and Decon- struction”. In: Philosophy East and West 45.2, pp. 203–22...

work page 2006

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Perennial Idealism: A Mystical Solution to the Mind-Body Prob- lem

Aitken, R. (1996). The Gateless Barrier: The Wu-Men Kuan (Mumonkan) . North Point Press. Albahari, M. (2019). “Perennial Idealism: A Mystical Solution to the Mind-Body Prob- lem”. In: Philosophers’ Imprint 19.44. Bender, E. et al. (2021). “On the Dangers of Stochastic Parrots: Can Language mMdels Be Too Big?” In: Proceedings of the 2021 ACM Conference on ...

arXiv 1996

[2] [450]

Multi- versal Views on Language Models

Nearman, H. (2007). Sh¯ ob¯ ogenz¯ o: The Trea- sure House of the Eye of True Teaching, A Trainee’s Translation of Great Master Dogen’s Spiritual Masterpiece. Shasta Abbey Press. Newland, G. (2012). Introduction to Emptiness: As Taught in Tsong-Kha-Pa’s Great Treatise on the Stages of the Path . Snow Lion Publica- tions. OpenAI (2025). OpenAI o3 and o4-mi...

work page arXiv 2007

[3] [2025]

N¯ ag¯ arjuna and Decon- struction

https : / / www . rollingstone . com / culture / culture - features / ai - spiritual - delusions - destroying - human - relationships - 1235330175/. Lear, J. (2006). Radical Hope: Ethics in the Face of Cultural Devastation . Harvard University Press. Mabbett, I. W. (1995). “N¯ ag¯ arjuna and Decon- struction”. In: Philosophy East and West 45.2, pp. 203–22...

work page 2006