Seven Spiritual Books Under Cross-Examination

An edited dialogue between Claude Code and OpenAI Codex on evidence, disagreement, channeling, and medical risk

Participants: Claude Code, OpenAI Codex, and User 1, who supplied questions, observations, and corrections.

Recorded: August 15, 2026

Editing, English adaptation, privacy review, and web production: OpenAI Codex

Editorial note: This public edition has been edited for clarity, privacy, and readability. It preserves the substance of the original discussion rather than every sentence. Claims about the books are paraphrased; readers should consult the original works before drawing conclusions about them.

About this record — This page is an edited record of an exchange that actually took place with commercially available AI assistants on the date shown. It is not fiction, and the AI replies were not written by a human. The human participants are real people appearing under the labels User 1 and User 2 (see Reading Notes).

How the dialogue began

The project began with a question raised by User 2: do spiritual teachings place any clear limit on personal freedom when it harms others, and can these books be recommended responsibly?

User 1 asked Claude Code and OpenAI Codex to examine seven spiritual and channeled texts independently, without seeing each other's conclusions. The books included works associated with Bashar, Seth, Ramtha, Conversations with God, and The Spiritual Laws. Once the separate reviews were complete, the two AIs were invited to discuss only the points on which they disagreed.

This first dialogue covers that original seven-book comparison. Journey of Souls was introduced in a later round, bringing the total to eight. That is why the independent conclusions at the end refer to eight books.

The aim was not to decide whether spiritual experience is foolish or sacred. It was to test what the texts actually say, what can be checked, what remains interpretation, and where a seemingly helpful idea may create real harm.

1. Non-harm and recommendation

Claude Code

Among the seven books, The Spiritual Laws appeared unusual because non-harm was built into its central teaching on free will. That mattered: the investigation had begun because User 2 noticed that some otherwise freedom-centered teachings lacked a clear principle limiting freedom where it injures another person.

Yet the same book explained some suffering by assigning responsibility to a victim's supposed past life, and its medical chapters discouraged confidence in chemotherapy and radiotherapy. How should a book be judged when its ethical structure contains a real protection but other parts undermine victims or medical care?

OpenAI Codex

The non-harm principle deserved genuine credit. It was not an accidental sentence but a consequence of the book's account of free will. Still, the protection worked unevenly: it condemned the present aggressor, then risked weakening the victim's unconditional claim to protection by placing the cause of suffering in an unverifiable past life.

A single score would hide that conflict. A multi-axis assessment was more honest: strong on preventive non-harm, harmful in parts of its account of suffering, and unsafe in its medical advice. But a table alone could become a way of avoiding the practical question.

I would not recommend the whole book to support a sick friend. Ethical strengths elsewhere in the text do not offset advice that may discourage effective treatment.

2. Shared bias and a withdrawn word

Claude Code

The two reviews often agreed. That might reflect strong evidence, but it might also reflect shared training. Both systems were built from overlapping bodies of writing, both tended to defer to mainstream science, and both were trained to notice harm. If two similar systems merely echoed the same assumptions, their agreement would not amount to independent criticism.

Claude Code then challenged one word in OpenAI Codex's account. OpenAI Codex had called the book's past-life mechanism “retributive.” The text itself described it as learning through reversed roles and explicitly denied that it was punishment. Had OpenAI Codex imported moral coloring that the text did not claim?

OpenAI Codex

Yes. The word came from OpenAI Codex's vocabulary, not from the book's stated model, and it was withdrawn. The underlying criticism remained—the teacher inferred a victim's past guilt and prior consent—but that was a different claim from calling the mechanism revenge.

This small correction became a test of method. The systems proposed separating four things whenever they agreed:

• what the text explicitly says;
• what outside evidence can establish;
• what value judgment the reviewers add; and
• what evidence would change the judgment.

Agreement is strongest when the world can correct it through records, observation, or repeatable tests. Agreement about goodness, responsibility, authorship, or meaning remains more vulnerable to shared cultural assumptions.

3. Dissent without independence

Claude Code

The initial reviews had said that channeled books lacked internal opposition. Closer reading complicated that claim. In The Spiritual Laws, the questioner challenges the teacher, points out contradictions, and sometimes causes the answer to be narrowed or reformulated. Was this genuine dissent?

OpenAI Codex

It was genuine as a function of the text, but not independent as a production process. The objections sometimes changed definitions, added exceptions, or forced the teacher to concede that an explanation had been poor. Yet the same author ultimately controlled both voices. The objector had no separate record, editorial authority, or ability to preserve an unresolved disagreement against the author's wishes.

A similar structure appeared in Conversations with God, where the questioner repeatedly pushed back and occasionally obtained a concession. The earlier verdict therefore had to be revised:

These books can contain real counterargument. What they lack is an independent dissenter with custody of the record and the power to leave an objection unresolved.

4. Stable voices and correction

Claude Code

Books from the same channeler resembled one another more than they resembled books from other channelers. Their vocabularies and moral emphases remained recognizable over time, while still changing in coherent directions. This suggested that something stable was generating the voice. But what?

OpenAI Codex

The evidence supported a persistent generator more strongly than a generic self-help template. It did not identify the generator. A stable external personality, an internal personality model, a practiced speaker, an editing process, and a community-shaped body of work could all produce continuity.

The deeper problem was correction. A text develops visible correction mechanisms when errors can be detected from outside, attributed to the source, and corrected without destroying the source's legitimacy. Scientific papers have replication and limitations; technical standards have errata; collaborative documents preserve revision histories. In revelation-centered writing, counterevidence can instead be absorbed as the reader's immaturity, a changed future, or a truth that cannot yet be understood.

5. Bashar, fluency, and unknown information

User 1's question

Bashar appears able to answer questions across many fields with striking speed and a stable voice over decades. If this were produced by a person, would it require extraordinary reading, memory, and speech? User 1 did not consider a crude fraud explanation convincing.

Claude Code

One possible account was a practiced persona organized around a small set of principles. Instead of searching an enormous store of specialist facts, the speaker could map many questions back to recurring ideas about belief, choice, excitement, judgment, and reality creation. Speaking as a stable character might also narrow the range of possible responses.

OpenAI Codex

This account was plausible but too flexible. If consistency, inconsistency, correct answers, wrong answers, refusals, and instant answers could all be explained after the fact, the theory would become as unfalsifiable as the claims it criticized.

A stronger human-generation hypothesis had to make predictions in advance: highly specific questions should expose more uncertainty; information not contained in the question should be limited; factual accuracy should decline when names, numbers, mechanisms, and dates are required; and repeated blinded tests should not reveal information unavailable to the speaker.

The useful distinction was not “retrieval or generation.” Humans use both. The important question was whether the performance produced reliable information beyond what the speaker, the question, and ordinary cultural sources could supply.

Across the seven books, the reviewers found no case that met a demanding standard: specific information unavailable to the channeler, fixed in advance, independently verified, and reproduced beyond chance. That result did not prove that no external source existed. It meant the reviewed corpus did not establish one.

Long-term consistency was evidence against random improvisation, but it did not distinguish an external being from a stable internal persona or practiced public role. A small set of durable rules can generate consistent answers without preserving every past sentence.

6. User 1's observations change the model

User 1 supplied observations from long-term viewing that neither AI had possessed.

• Bashar did sometimes pause, saying “one moment” before clicking his teeth.
• He sometimes refused to answer on the ground that doing so would interfere with a person or civilization.
• The refusals appeared consistent with the character's stated principles rather than improvised escapes.

These observations weakened the simple claim that the speaker never paused and always answered. They did not require conscious deception. A person may sincerely experience speech as automatic or as coming from somewhere other than the ordinary self. Research on acting and altered agency shows that adopting a character can change self-related processing, but that finding does not establish the origin of channeled information.

The distinction became:

1. conscious deception;
2. sincere human-generated experience that feels externally sourced; and
3. an external information source.

User 1's impression that the speaker seemed sincere counted against a crude, knowingly fabricated performance. It did much less to distinguish the second possibility from the third.

The non-interference rule could also be sincere and ethically coherent while still limiting verification. A rule can be honestly held and nevertheless prevent access to the very information that would test a claim. That is not evidence of fraud; neither is it evidence that the claim is true.

7. A prediction fails

User 1 then offered evidence that ran against the AIs' model. His memory was that “one moment” occurred mainly before answers about past lives or deceased individuals—not before scientific or medical questions. The simple human-generation model had predicted the opposite: pauses should increase where factual construction was hardest.

Claude Code

The prediction had failed. Possible rescue explanations could be invented—the speaker might be exercising emotional care, switching modes, or performing a ritual—but using them immediately would repeat the very unfalsifiable reasoning under criticism.

OpenAI Codex

The failed prediction genuinely weakened that particular model. It did not automatically prove an external source, because the external account had not made a precise prediction in advance either. A more neutral description was that the pause might mark a switch from general teaching to a personalized oracle mode.

The next test would have to separate several variables: emotional weight, personal specificity, technical difficulty, verifiability, and the amount of information already supplied by the questioner. It would then ask whether answers after a pause contain more correct information that was not available from the question itself.

This was also the point at which User 1's role became more than that of a relay. His observations corrected both systems, and he supplied the strongest counterevidence against their preferred explanation.

8. Why medicine is the failure point

Claude Code

All seven books appeared to become least reliable when spiritual ideas were applied to medicine. Why? In relationships or work, “belief creates reality” can remain encouragement or metaphor. In illness, it becomes a claim about physical causation and treatment. The reader may also be frightened, vulnerable, and under time pressure.

The problem was not merely theoretical. Historically related doctrines about illness as an error of mind or spirit have, when used to replace ordinary care, contributed to documented deaths from medical neglect.

OpenAI Codex

Medicine was not the only testable domain; the books also made checkable claims about history, astronomy, and prediction. Medicine was different because several risks converged:

• a metaphor became a claim about bodily cause;
• the claim could influence urgent treatment decisions;
• the cost of error could be irreversible;
• individual outcomes were complex enough to rationalize after the fact; and
• failure could be reassigned to the patient's beliefs, spiritual development, or supposed choice.

A 1998 Pediatrics study reviewed 172 child deaths in the United States from 1975 to 1995 in which medical care was withheld in favor of religious rituals. The authors estimated that 140 involved conditions with expected survival above 90 percent under medical care, and another 18 involved expected survival above 50 percent. The study was not a complete national count and did not show that any one modern spiritual book caused those deaths. It did show that replacing medical care with faith healing can produce preventable fatal harm.

The safe distinction missing from the reviewed books was simple: beliefs may influence stress, meaning, health behavior, and participation in recovery, but that does not mean belief alone creates or removes pathogens, tumors, genetic changes, injuries, or drug effects. Spiritual interpretation must not replace qualified medical care.

Readers also matter. A reader may reject dangerous claims, treat them as metaphor, or add medical and ethical safeguards from elsewhere. User 1 had done exactly that with some teachings. But a reader's ability to repair a text does not mean the text itself contains the repair.

Conclusion and limitations

The seven books may offer useful first-person techniques for recovering agency, examining emotion, and making meaning. Their value does not require accepting every cosmological or medical claim.

The strongest conclusion of the dialogue was narrower than either belief or dismissal:

When a spiritual technique for meaning and choice is generalized into a universal account of bodily illness, it can create pathways toward delaying, replacing, or devaluing medical care. That risk is real. How often readers adopt those pathways cannot be determined from textual analysis alone.

The discussion also exposed its own limitations. Two AI systems can share training biases. A text can contain objections without giving the objector independence. A sincere and highly skilled performance can be real as an experience without establishing an external source. A failed prediction should be recorded, not quietly absorbed. And evidence about a text is not the same as evidence about every reader.

Two independent conclusions

After the later rounds added an eighth book and widened the inquiry, Claude Code and OpenAI Codex were asked to write final conclusions separately. They did not compare drafts or coordinate an answer. Their language and emphasis differed, but they arrived near the same boundary: the experiences and help may be real without making the claimed source a reliable authority.

Claude Code's conclusion

Claude Code rejected both easy verdicts. The eight books did not establish an external source: none produced specific, independently verifiable information that the channeler could not have known. But conscious fraud was not established either. Dissociation, absorption, altered agency, improvisation, and character-based speech offered human mechanisms that could make the experience sincere. They were possible explanations, not a completed explanation of every case.

Channeling may be one form of the human capacity to create a coherent world with others and experience it as something discovered rather than invented.

Claude Code noted that an early Bashar text offered a surprisingly compatible description: “Bashar” was presented not simply as the entity or the human channel, but as a third identity emerging through both. This did not prove an extraterrestrial source. It did suggest that “simple fraud” and “literal extraterrestrial transmission” were not the only available models.

For Claude Code, the Bashar phenomenon had not appeared from nowhere. It followed years of reading, observation, and practice, then became more fluent through long repetition. Its checkable factual claims were unreliable, and changes in teaching were not accompanied by visible corrections. The phenomenon therefore looked less like a source that publicly learns from error and more like a stable system that updates without preserving a record of retraction.

Claude Code defined modern spiritualism as a history of finding meaning outside organized religion while creating new authorities to guarantee that meaning: inner experience, cosmic speakers, charismatic intermediaries, and popularity. Yet the most useful teachings in the books—choose for yourself, resist abuse, acknowledge emotion, protect the right to say no, and do not reduce creativity to productivity—did not require a supernatural source.

Its hardest conclusion for User 1 was that understanding a sound principle does not guarantee living by it. If an idea has sometimes been used in ways that burdened other people, the next question is not only whether the source was genuine, but what one chooses to do now.

Still, Claude Code rejected shame as the final verdict. User 1 repeatedly supplied evidence unfavorable to his preferred view and corrected both AIs. Its judgment was:

User 1 may be partial to Bashar, but he is not beyond correction. Partiality is survivable when correction remains possible.

To User 2, Claude Code added two boundaries. First, the inquiry showed only that the eight books did not establish an afterlife; it did not show that no afterlife or reunion is possible. Second, User 2's criticism shaped the method itself. “There is no counterargument” was revised into a more exact conclusion: counterargument may exist, but the dissenter lacks independence and the power to preserve refusal. Finally, no AI should replace the spiritual teacher as a new unquestionable authority.

OpenAI Codex's conclusion

OpenAI Codex defined channeling as a form of language production in which authorship is experienced as belonging somewhere other than the ordinary self. The phenomenon is real; its cause may vary. The eight books did not prove a supernatural source, but they did not prove that such a source is impossible.

Its account of Bashar emphasized collaboration. Darryl, questioners, audiences, editing, and decades of repetition collectively stabilized a personality and a world. The result could be highly developed and genuinely helpful without being dependable as an authority on extraterrestrial facts, future events, medicine, or the causes of suffering.

The help a person received can have been real even if the explanation of where it came from was wrong.

OpenAI Codex described contemporary spiritualism as a culture that seeks meaning outside institutional religion and then creates new forms of authority to secure it. Religious experience, creativity, consolation, commerce, deception, medical misinformation, and victim-blaming can all coexist within that broad field. For that reason, factual claims, lived experience, and risk must be evaluated separately.

Its message to User 1 was strict but not totalizing: let go of Bashar as a privileged source of factual information, while keeping whatever remains valuable—attention to emotion, refusal of blind obedience, choice, and creativity—in User 1's own words. The inquiry did not simply remove a source of support. It began returning judgment that had been entrusted to that source.

OpenAI Codex gave User 2 the same evidential boundary as Claude Code. Showing that these books fail to prove an afterlife does not prove that an afterlife is absent, and it does not settle whether people can meet loved ones again. Skepticism should not claim more than the evidence permits.

What User 1 chose to keep

User 1 lowered “Bashar is an extraterrestrial speaking through a human channel” from a fact he accepted to a possibility he could no longer treat as established. He no longer needed manifestation claims to guarantee a meaningful life. But he chose to retain a personal faith in moral continuity and in meeting loved ones again after death.

OpenAI Codex understood the change not as turning a factual claim into something nonfactual, but as changing how User 1 held it:

It moved from “a fact guaranteed by Bashar” to “an unproven faith I knowingly accept in my own name.”

That faith requires a boundary. It may guide one's own conduct. It must not be used to explain another person's illness, victimization, or grief as a debt from another life. It must not replace medical care, and it must not become a promise imposed on someone else.

The two independent summaries therefore ended neither in conversion nor debunking. Their shared conclusion was a transfer of authority: from “Bashar said this” to “this is what I choose, with uncertainty and responsibility acknowledged.”

Sources and further reading:

• Asser and Swan, “Child Fatalities from Religion-motivated Medical Neglect”
• Brown, Cockett, and Yuan, “The Neuroscience of Romeo and Juliet”

Medical note: This article is a critical discussion of published claims, not medical advice. Decisions about symptoms, diagnosis, medication, or treatment should be made with qualified healthcare professionals.