BREW-AIDER-CVE-2026-96636 (GHSA-P8HX-V2P6-RP84)
Vulnerability from osv_homebrew – Published: 2026-10-29 21:58 – Updated: 2026-10-29 21:58 – Source websiteSummary
Pillow's decompression-bomb protection (Image._decompression_bomb_check, gated by Image.MAX_IMAGE_PIXELS) is enforced only on the first frame, inside Image.open(). For multi-frame formats, seek() re-parses the target frame's header and resets self._size. GifImagePlugin and TiffImagePlugin re-run the bomb check on seek(), but MpoImageFile.seek() and DcxImageFile.seek() do not. A crafted MPO or DCX whose first frame is tiny (passing the check) but whose later frame declares enormous dimensions therefore bypasses the protection: loading that frame decodes an over-limit image and commits large amounts of memory — a denial of service (memory exhaustion / OOM).
Same class as the recently assigned CVE-2026-55379 (BdfFontFile), CVE-2026-55380 (GdImageFile) and CVE-2026-40192 (FITS GZIP).
Details
Image.open() → _open_core() runs _decompression_bomb_check(im.size) once, on the frame present at open time. Both vulnerable seek() methods re-parse a new frame and set a new self._size without re-checking:
PIL/MpoImagePlugin.py,MpoImageFile.seek(): callsJpegImagePlugin.JpegImageFile._open(self)(which setsself._sizefrom the frame's JPEG SOF) and rebuildsself.tile, with noImage._decompression_bomb_check(self.size).PIL/DcxImagePlugin.py,DcxImageFile.seek(): callsPcxImageFile._open(self)(which setsself._sizefrom the page's PCX header), with no check.
Confirmed present in 12.3.0 (latest release) and on main (13.0.0.dev0). Compare GifImagePlugin._seek and TiffImagePlugin, which call _decompression_bomb_check correctly on seek.
PoC
Self-contained; the victim uses default MAX_IMAGE_PIXELS:
import io
from PIL import Image
# attacker builds the file (guard disabled only to CREATE the oversized frame)
Image.MAX_IMAGE_PIXELS = None
small = Image.new("RGB", (8, 8))
big = Image.new("RGB", (14000, 14000), (120, 60, 30)) # 196,000,000 px > error threshold
buf = io.BytesIO()
small.save(buf, format="MPO", save_all=True, append_images=[big])
data = buf.getvalue() # ~3 MB
Image.MAX_IMAGE_PIXELS = int(1024 * 1024 * 1024 // 4 // 3) # restore default
# victim (default limits)
im = Image.open(io.BytesIO(data)) # frame 0 (8x8) passes the bomb check
im.seek(1) # frame 1 (14000x14000) -> NOT re-checked <-- bug
im.load() # decodes the over-limit image, commits memory
print("loaded over-limit frame:", im.size)
Measured on Pillow 12.3.0 (default limit 89,478,485 px; error threshold 178,956,970 px):
| Vector | File size | Frame 1 | Peak RSS |
|---|---|---|---|
| MPO (JPEG) | ~3.0 MB | 14000×14000 (196 Mpx) | ~763 MB |
| DCX (PCX) | ~6.2 MB | 14000×14000 (196 Mpx) | ~200 MB |
| Control: same image standalone | — | 196 Mpx | DecompressionBombError (correctly blocked) |
The DCX vector (0x3ADE68B1 magic + offset directory + two PCX pages) behaves identically via DcxImageFile.seek(). Scaling frame 1 toward JPEG's ~65500×65500 maximum drives multi-GB commitments from a small file, reliably OOM-killing the process.
Impact
Denial of service (memory exhaustion / OOM) — CWE-400 / CWE-1284, a bypass of Pillow's documented decompression-bomb mitigation. seek() + load() is the normal way to iterate frames (e.g. PIL.ImageSequence.Iterator, thumbnailers, converters, 3D-photo handlers). Because MPO is auto-detected from a JPEG carrying MPF metadata, a file served with a .jpg extension can be opened as MPO, so any application that iterates frames of untrusted images is exposed.
Suggested fix
Re-run the check after seek() re-parses a frame, mirroring GIF/TIFF:
# PIL/MpoImagePlugin.py — end of MpoImageFile.seek()
self.tile = [ImageFile._Tile("jpeg", (0, 0) + self.size, self.offset, self.tile[0][-1])]
Image._decompression_bomb_check(self.size) # add
self.__frame = frame
# PIL/DcxImagePlugin.py — end of DcxImageFile.seek()
PcxImageFile._open(self)
Image._decompression_bomb_check(self.size) # add
{
"affected": [
{
"ecosystem_specific": {
"fix": null,
"range_state": "affected",
"resource": "pillow",
"resource_purl": "pkg:pypi/pillow@12.1.1",
"upstream_fixed_in": "13.0.0"
},
"package": {
"ecosystem": "Homebrew",
"name": "aider",
"purl": "pkg:brew/aider"
},
"ranges": [
{
"events": [
{
"introduced": "0.86.1_1"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"database_specific": {
"confidence": "high",
"source": "matched",
"strategy": "registry",
"upstream_evidence": [
{
"ecosystem": "PyPI",
"key": "pkg:pypi/pillow@12.1.1",
"name": "pillow",
"resource": "pillow",
"strategy": "registry",
"subject_version": "12.1.1"
}
]
},
"details": "### Summary\nPillow\u0027s decompression-bomb protection (`Image._decompression_bomb_check`, gated by `Image.MAX_IMAGE_PIXELS`) is enforced only on the **first frame**, inside `Image.open()`. For multi-frame formats, `seek()` re-parses the target frame\u0027s header and resets `self._size`. `GifImagePlugin` and `TiffImagePlugin` re-run the bomb check on `seek()`, but **`MpoImageFile.seek()` and `DcxImageFile.seek()` do not.** A crafted MPO or DCX whose first frame is tiny (passing the check) but whose later frame declares enormous dimensions therefore bypasses the protection: loading that frame decodes an over-limit image and commits large amounts of memory \u2014 a denial of service (memory exhaustion / OOM).\n\nSame class as the recently assigned CVE-2026-55379 (BdfFontFile), CVE-2026-55380 (GdImageFile) and CVE-2026-40192 (FITS GZIP).\n\n### Details\n`Image.open()` \u2192 `_open_core()` runs `_decompression_bomb_check(im.size)` once, on the frame present at open time. Both vulnerable `seek()` methods re-parse a new frame and set a new `self._size` without re-checking:\n\n- **`PIL/MpoImagePlugin.py`, `MpoImageFile.seek()`**: calls `JpegImagePlugin.JpegImageFile._open(self)` (which sets `self._size` from the frame\u0027s JPEG SOF) and rebuilds `self.tile`, with no `Image._decompression_bomb_check(self.size)`.\n- **`PIL/DcxImagePlugin.py`, `DcxImageFile.seek()`**: calls `PcxImageFile._open(self)` (which sets `self._size` from the page\u0027s PCX header), with no check.\n\nConfirmed present in **12.3.0** (latest release) and on **`main`** (13.0.0.dev0). Compare `GifImagePlugin._seek` and `TiffImagePlugin`, which call `_decompression_bomb_check` correctly on seek.\n\n### PoC\nSelf-contained; the victim uses **default** `MAX_IMAGE_PIXELS`:\n\n```python\nimport io\nfrom PIL import Image\n\n# attacker builds the file (guard disabled only to CREATE the oversized frame)\nImage.MAX_IMAGE_PIXELS = None\nsmall = Image.new(\"RGB\", (8, 8))\nbig = Image.new(\"RGB\", (14000, 14000), (120, 60, 30)) # 196,000,000 px \u003e error threshold\nbuf = io.BytesIO()\nsmall.save(buf, format=\"MPO\", save_all=True, append_images=[big])\ndata = buf.getvalue() # ~3 MB\nImage.MAX_IMAGE_PIXELS = int(1024 * 1024 * 1024 // 4 // 3) # restore default\n\n# victim (default limits)\nim = Image.open(io.BytesIO(data)) # frame 0 (8x8) passes the bomb check\nim.seek(1) # frame 1 (14000x14000) -\u003e NOT re-checked \u003c-- bug\nim.load() # decodes the over-limit image, commits memory\nprint(\"loaded over-limit frame:\", im.size)\n```\n\nMeasured on Pillow 12.3.0 (default limit 89,478,485 px; error threshold 178,956,970 px):\n\n| Vector | File size | Frame 1 | Peak RSS |\n|---|---|---|---|\n| MPO (JPEG) | ~3.0 MB | 14000\u00d714000 (196 Mpx) | ~763 MB |\n| DCX (PCX) | ~6.2 MB | 14000\u00d714000 (196 Mpx) | ~200 MB |\n| Control: same image standalone | \u2014 | 196 Mpx | `DecompressionBombError` (correctly blocked) |\n\nThe DCX vector (`0x3ADE68B1` magic + offset directory + two PCX pages) behaves identically via `DcxImageFile.seek()`. Scaling frame 1 toward JPEG\u0027s ~65500\u00d765500 maximum drives multi-GB commitments from a small file, reliably OOM-killing the process.\n\n### Impact\nDenial of service (memory exhaustion / OOM) \u2014 CWE-400 / CWE-1284, a bypass of Pillow\u0027s documented decompression-bomb mitigation. `seek()` + `load()` is the normal way to iterate frames (e.g. `PIL.ImageSequence.Iterator`, thumbnailers, converters, 3D-photo handlers). Because **MPO is auto-detected from a JPEG carrying MPF metadata**, a file served with a `.jpg` extension can be opened as MPO, so any application that iterates frames of untrusted images is exposed.\n\n### Suggested fix\nRe-run the check after `seek()` re-parses a frame, mirroring GIF/TIFF:\n\n```python\n# PIL/MpoImagePlugin.py \u2014 end of MpoImageFile.seek()\nself.tile = [ImageFile._Tile(\"jpeg\", (0, 0) + self.size, self.offset, self.tile[0][-1])]\nImage._decompression_bomb_check(self.size) # add\nself.__frame = frame\n\n# PIL/DcxImagePlugin.py \u2014 end of DcxImageFile.seek()\nPcxImageFile._open(self)\nImage._decompression_bomb_check(self.size) # add\n```",
"id": "BREW-aider-CVE-2026-96636",
"modified": "2026-10-29T21:58:11Z",
"published": "2026-10-29T21:58:11Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/python-pillow/Pillow/security/advisories/GHSA-p8hx-v2p6-rp84"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-96636"
},
{
"type": "WEB",
"url": "https://github.com/python-pillow/Pillow/pull/10011"
},
{
"type": "WEB",
"url": "https://github.com/python-pillow/Pillow/commit/f1acfc54c53502948d0d14121872061df931db48"
},
{
"type": "PACKAGE",
"url": "https://github.com/python-pillow/Pillow"
},
{
"type": "WEB",
"url": "https://github.com/python-pillow/Pillow/releases/tag/13.0.0"
}
],
"schema_version": "1.7.3",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
],
"summary": "Pillow: Decompression-bomb protection bypass via multi-frame seek() in MPO and DCX (MpoImageFile.seek / DcxImageFile.seek)",
"upstream": [
"GHSA-p8hx-v2p6-rp84",
"BIT-pillow-2026-96636",
"CVE-2026-96636",
"PYSEC-2026-4337"
]
}
Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.
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.
Vulnerability timeline
Sightings and recorded dates, grouped by day · UTC
How durations are calculated
Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.
Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.
Loading timeline…
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.