Provenance Record

Clean 571-word Briefing (range 300-800) with Overview, What We Know, earlier-fix context, What We Don't Know and a short Analysis. Neutral, actionable, no AI self-reference.

Read all three snapshots from disk (gunzip). source-0.html.gz (docs.gitlab.com 19.4.1 notice, HTTP 200): confirms CVE-2026-90970; title 'Improper Neutralization issue in custom flow prompt template impacts AI Gateway'; severity Critical; CVSS 9.9 with vector CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H; impacted versions 'all versions from 18.1.6 before 19.2.4, 19.3 before 19.3.2, and 19.4 before 19.4.1'; fixed 19.2.4, 19.3.2, 19.4.1; credit 'Thanks invisiblemeerkat for responsibly disclosing this issue'; the quoted sentence ('escape the prompt template sandbox via a specially crafted flow configuration, leading to arbitrary command execution on the AI Gateway') is verbatim; 'targeted outreach ... prior to this release post' and GitLab-hosted gateway/GitLab.com/Dedicated 'do not need to take action' confirmed. The snapshot carries no publication date, as the article states. source-1.html.gz (18.8.1 notice, HTTP 200): 'On February 6, 2026' release of 18.6.2, 18.7.1, 18.8.1; CVE-2026-1868 'Insecure Template expansion', Duo Workflow Service component, crafted Duo Agent Platform Flow definitions, authenticated access required, 'Denial of Service or gain code execution on the Gateway'; CVSS 9.9 with the identical vector string; 'discovered internally by GitLab team member Joern Schneeweisz'. Neither notice states any relationship between the two CVEs; the article says so explicitly and implies none. source-2.html.gz (BleepingComputer): fetched via Archive.org fallback but is NOT a challenge page; it is the full article (Wayback capture 02 Oct 2026) with byline 'By Sergiu Gatlan October 2, 2026 12:20 PM', quoting the same GitLab advisory text. The October 2, 2026 report date is therefore verified against a cited source and needs no removal. suspicious_patterns null for all three sources. Domains docs.gitlab.com and bleepingcomputer.com are on config/source_allowlist.txt.

CVE ID, CVSS score and vector, affected and fixed versions, credit, February date and CVE-2026-1868 details all match the snapshots verbatim. The flaw is described only in GitLab's wording (prompt template sandbox escape; arbitrary command execution on the AI Gateway by an authenticated Duo Agent Platform user); the body never claims 'remote code execution' (that phrase appears only in the BleepingComputer URL/headline and the 'remote-code-execution' tag). The 'no known exploitation' handling is accurate: GitLab's notice says nothing about exploitation and BleepingComputer reports none for this CVE (its CISA exploitation mention concerns a different CVE-2026-85706); the article only says the notice does not say and the reviewed sources do not report it. The 'same authenticated-access precondition' and 'flow definitions or flow configuration' comparison is supported. No version-lineage claims beyond quoted ranges. Internal link to the Bifrost article resolves to an existing published article.

Accurate, fully sourced Cybersecurity Briefing; all dates, identifiers and quotes verified against snapshots. Approve.