Cannabinoids and cancer across humans, dogs and cats

What the test completed

Six memos were saved on 26 September 2026 (UTC). The Chief Scientist’s memo came from the earlier research lab. Five further roles produced memos with Gemini 2.5 Flash. Later roles could read earlier conclusions, so agreement between them cannot count as independent replication.

[ARCH]

What did the human studies measure?

The ASCO guideline recommends against using cannabis or cannabinoids as cancer-directed treatment outside a clinical trial. It finds a narrower possible role alongside standard antiemetics for refractory chemotherapy-induced nausea and vomiting in adults. This does not justify a general claim that cannabinoids treat cancer pain.

In Segar and colleagues’ single-arm dronabinol study in people with breast cancer and bone metastases, 4 of 14 evaluable patients reduced opioid use by at least 20%. Without a comparison group it cannot establish how much of that change the drug caused. The endpoint was opioid use, not tumour response; dronabinol is not CBD.

[ASCO] [PILOT]

Which compounds were studied in dogs?

A canine lymphoma study investigated endocannabinoids in the laboratory. Endocannabinoids made by the body are distinct from plant-derived CBD. In dogs with lymphoma receiving chemotherapy, another study examined CBD/CBDA-rich oil, including safety and doxorubicin exposure. Neither observation demonstrates that CBD prolongs survival or treats lymphoma across species.

[DOGCELL] [DOGPK]

What did the search find in cats?

The September search did not establish a direct feline cannabinoid–cancer intervention in the material reviewed. That describes a limited search. It does not establish that no such study exists, that cannabinoids have no effect, or that a positive effect in another species transfers to cats. The archival scientist code used a bounded, prioritised sample and one hard-coded Scout run plus a stored gap record.

[ARCH]

What the audit changed

One memo lists PMID 21993992 among animal patient studies relevant to its canine cancer discussion. That identifier belongs to an article on ultrasound in horses. The citation register records the mismatch beside the original reference.

The five new memos display May 2024 dates. The stored completion records date all five to September 2026. “4 of 132” counts labelled reports in a selected collection. It cannot be interpreted as a patient response rate.

Two memos also cite the RESONANCE protocol alongside evidence of symptom relief. A protocol documents a study plan; it does not report an efficacy result. Bibliographic identity and support for a particular claim are checked separately.

[ARCH] [MISMATCH] [PROTOCOL]

What would justify another conclusion?

For each proposed claim, freeze the input packet and search criteria, link the exact paper and endpoint, and obtain a review independent of its author. In a treatment study, specify species, comparator, clinical endpoint and uncertainty. Report null or conflicting results too. Project research memos and proposals may be published with status and provenance; only an independent verification process can promote evidence.

Sources for the digest

1. Braun et al. (2024), ASCO guideline on cannabis and cannabinoids in adults with cancer. guideline, external check
2. Segar et al. (2026), dronabinol and opioid use in cancer-induced bone pain. primary human study
3. Omer et al. (2023), endocannabinoids and canine lymphoma. primary laboratory study
4. CBD/CBDA-rich hemp oil and doxorubicin pharmacokinetics in dogs receiving lymphoma chemotherapy. primary canine patient study
5. Barton (2011), abdominal ultrasonography in horses. Rejected as support for the Sceptic’s claim. citation mismatch
6. Original six-row BigQuery export: generation metadata and unedited memo text. historical agent output, not primary biological evidence
7. RESONANCE trial protocol, advanced cancer symptom burden: study design, not outcome data. protocol, not a clinical result

Six original memos

These archived outputs preserve what each role said at the time, including mistakes. The correction notes above are part of the public record.

• Chief Scientist
• Cross Species Analyst
• Gap Finder
• Sceptic
• Experiment Designer
• Translator

Citation register

All 34 PubMed identifiers mentioned in the memos are listed. The review column shows which claims have been checked. The source packets supplied to each role were not preserved.

Research data

• Research note (JSON)
• Original memo export (JSON)
• Citation register (JSON)