GHSA-M29V-3G7G-9C63
Vulnerability from github – Published: 2026-11-19 16:00 – Updated: 2026-11-19 16:00What's wrong
search_claims never calls _row_verified_on_read. Its row loop applies only the PRELIMINARY-unenrolled filter, then appends the row. query_claims runs the same projection but additionally re-verifies REPLICATED/ESTABLISHED signatures on read (line 4023) and drops any row that fails, per the documented tamper guarantee in the comment block above it: a high-trust row whose signature does not verify is excluded from query results entirely.
So a tampered or forged high-trust row splits the two read paths. Any process with DB write access, or an attacker copying a genuine validation envelope onto a fresh row, can flip a row to ESTABLISHED. query() drops it; search() serves it at its persisted trust level. Verified in a runtime check against a forged ESTABLISHED row:
query() returns forged row: False
search() returns forged row: True # support_level ESTABLISHED, 'verified' key absent
The docstring makes this worse: search_claims promises "the same projection as query_claims" and that the include_unverified filter "applies identically", so callers reasonably trust search results equally. The forged row also lacks the verified key entirely, so a consumer cannot tell it went unchecked.
Severity and impact
High. The verify-on-read tamper exclusion is a stated trust guarantee, and search is a primary read surface. A forged ESTABLISHED finding that query correctly rejects is served by search as genuine, with no marker that verification was skipped.
Location
mareforma/db/core.py:4206search_claims row loop applies only the PRELIMINARY-unenrolled filter, then appendsmareforma/db/core.py:4114search_claims signature and docstring, which claims query_claims-identical semanticsmareforma/db/core.py:4023query_claims verify-on-read call that search_claims lacks
Fix direction
Mirror query_claims: add a verify_cache: dict = {} and, inside the search_claims row loop after the include_unverified check, drop rows that fail _row_verified_on_read(conn, d, verify_cache). Since the verify filter can drop rows even when include_unverified=True, remove the single-batch break on include_unverified and keep batching until limit survivors or the FTS5 match-set is exhausted, with the same scan ceiling query_claims uses. Also set single_trust_domain / trust_domain_root on surviving ESTABLISHED rows if the "same projection" promise is to hold.
Do red/green TDD: add a regression test that fails on main, then make it pass.
Quality bar
SOLID, DRY, KISS. Clean, readable, well-tested code. No new errors, warnings, or test failures; the full suite stays green on all supported Python versions.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "mareforma"
},
"ranges": [
{
"events": [
{
"introduced": "0.3.7"
},
{
"fixed": "0.3.8"
}
],
"type": "ECOSYSTEM"
}
],
"versions": [
"0.3.7"
]
}
],
"aliases": [
"CVE-2026-71533"
],
"database_specific": {
"cwe_ids": [
"CWE-347"
],
"github_reviewed": true,
"github_reviewed_at": "2026-11-19T16:00:52Z",
"nvd_published_at": "2026-11-18T23:17:21Z",
"severity": "MODERATE"
},
"details": "## What\u0027s wrong\n\n`search_claims` never calls `_row_verified_on_read`. Its row loop applies only the PRELIMINARY-unenrolled filter, then appends the row. `query_claims` runs the same projection but additionally re-verifies REPLICATED/ESTABLISHED signatures on read (line 4023) and drops any row that fails, per the documented tamper guarantee in the comment block above it: a high-trust row whose signature does not verify is excluded from query results entirely.\n\nSo a tampered or forged high-trust row splits the two read paths. Any process with DB write access, or an attacker copying a genuine validation envelope onto a fresh row, can flip a row to ESTABLISHED. `query()` drops it; `search()` serves it at its persisted trust level. Verified in a runtime check against a forged ESTABLISHED row:\n\n```\nquery() returns forged row: False\nsearch() returns forged row: True # support_level ESTABLISHED, \u0027verified\u0027 key absent\n```\n\nThe docstring makes this worse: `search_claims` promises \"the same projection as query_claims\" and that the `include_unverified` filter \"applies identically\", so callers reasonably trust search results equally. The forged row also lacks the `verified` key entirely, so a consumer cannot tell it went unchecked.\n\n## Severity and impact\n\n**High.** The verify-on-read tamper exclusion is a stated trust guarantee, and search is a primary read surface. A forged ESTABLISHED finding that query correctly rejects is served by search as genuine, with no marker that verification was skipped.\n\n## Location\n\n- `mareforma/db/core.py:4206` search_claims row loop applies only the PRELIMINARY-unenrolled filter, then appends\n- `mareforma/db/core.py:4114` search_claims signature and docstring, which claims query_claims-identical semantics\n- `mareforma/db/core.py:4023` query_claims verify-on-read call that search_claims lacks\n\n## Fix direction\n\nMirror `query_claims`: add a `verify_cache: dict = {}` and, inside the search_claims row loop after the include_unverified check, drop rows that fail `_row_verified_on_read(conn, d, verify_cache)`. Since the verify filter can drop rows even when `include_unverified=True`, remove the single-batch break on `include_unverified` and keep batching until `limit` survivors or the FTS5 match-set is exhausted, with the same scan ceiling `query_claims` uses. Also set `single_trust_domain` / `trust_domain_root` on surviving ESTABLISHED rows if the \"same projection\" promise is to hold.\n\nDo red/green TDD: add a regression test that fails on `main`, then make it pass.\n\n## Quality bar\n\nSOLID, DRY, KISS. Clean, readable, well-tested code. No new errors, warnings, or test failures; the full suite stays green on all supported Python versions.",
"id": "GHSA-m29v-3g7g-9c63",
"modified": "2026-11-19T16:00:52Z",
"published": "2026-11-19T16:00:52Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/mareforma/mareforma/security/advisories/GHSA-m29v-3g7g-9c63"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-71533"
},
{
"type": "WEB",
"url": "https://github.com/mareforma/mareforma/pull/58"
},
{
"type": "WEB",
"url": "https://github.com/mareforma/mareforma/commit/8b5291792128326a22f240123659d2754446c194"
},
{
"type": "PACKAGE",
"url": "https://github.com/mareforma/mareforma"
},
{
"type": "WEB",
"url": "https://github.com/mareforma/mareforma/releases/tag/v0.3.8"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "Mareforma: search_claims skips verify-on-read, serving forged rows"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.