P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

Rails fixed an Active Storage flaw on 4 Nov; the CVE record said 9.8 that day, the Rails advisory came 12 Nov

Rails shipped the Active Storage fix on 4 November 2026, CVE.org carried a 9.8 record for CVE-2026-122006 that afternoon, and the Rails advisory followed on 12 November, 8 days 9 hours later (derived). A second CVE id was reserved and rejected in between; no saved page says why.

By · LinkedIn · · 20 min read

8 days 9 hours from the CVE record to the Rails advisory, and what that gap does not show

The Rails maintainers published Active Storage fix releases 8.0.5.2 on Wednesday 4 November 2026 at 02:31:54 UTC and 8.1.4.1 at 02:46:00 UTC. The changelog item for the fix carries the advisory id GHSA-f3w6-w445-fwh9. CVE.org published CVE-2026-122006 the same day at 14:12:17 UTC, assigned by MITRE, with a CVSS 3.1 base score of 9.8. The Rails advisory itself, "Active Storage: Possible arbitrary file write and read in image analysis", carries a published time of 12 November at 23:46:51 UTC. From the CVE record to the advisory is 8 days 9 hours 34 minutes (derived: 4 November 14:12:17 to 12 November 23:46:51 UTC is 201 hours 34 minutes 34 seconds). From the 8.1.4.1 release to the advisory it is 8 days 21 hours 0 minutes 51 seconds (derived).

What the gap does not establish. It does not say when anyone outside the project learned the details, because the CVE record of 4 November already lists the advisory's address among its references, and no saved page says whether that address answered then. It does not say why a 9.8 record existed first. And the sequence has a second strand: a different CVE id, CVE-2026-126082, was reserved and published on 10 November and rejected on 12 November as "a duplicate of another CVE", 7 hours 44 minutes before the advisory appeared (derived). The advisory still lists it. This briefing reports the order and leaves the cause open.

The release notes and the advisory are the Rails maintainers' own account of their own flaw; the saved pages state no commercial interest for them. A Turing Institute observatory report and a DigitalOcean pull request also touch this bug. They are third parties, not the vendor, and are set out separately. This site gives no verdict on any of them.

What the maintainers say they fixed

Both releases carry the same Active Storage changelog item, headed "Ship an ImageMagick security policy and apply it to every ImageMagick process MiniMagick spawns" and tagged GHSA-f3w6-w445-fwh9. Its account of the flaw is short: ImageMagick chooses a coder from the content of the file it is given, "Active Storage constrained none of ImageMagick's coders", and the ImageMagick analyzer runs when a file is attached. The advisory of 12 November says the same and adds "Attaching the upload is sufficient". The fix is a policy file that Active Storage now ships, added to ImageMagick's search path so that it applies to every ImageMagick command MiniMagick runs, not only Active Storage's.

Drawn from the Rails release notes of 4 November 2026 and the Rails advisory of 12 November 2026. No attack detail is drawn.

The notes list what the shipped policy denies, by coder family: scripting (MSL), reading or deleting files other than the image (TEXT, EPHEMERAL), fetching remote references (URL, HTTP, HTTPS, FTP), displaying to a screen (SHOW, WIN, PLT), everything handed to Ghostscript, and direct file access (FILE, MPC, XPS, PCL), plus @filename indirection. The notes then name who must act, which the diagram shows: mini_magick 5.1 or later, apps that render PostScript or PDF, and apps that never receive the shipped policy. They end with "Every other application needs no change".

Three things the changelog item does not contain, read in full in both releases: a CVE id, a severity, and any statement about the Debian and Ubuntu policies or uploads from untrusted users. The same notes carry 19 distinct advisory ids in 8.0.5.2 and 21 in 8.1.4.1 (derived by counting); the others belong to other fixes and are not covered here. So on 4 November the public text said what changed and who must act, while the CVE record said 9.8; the advisory's list of conditions came on 12 November.

Two CVE ids: the sequence the records give

The diagram and table set out every dated step in the saved records, in UTC. The first CVE record was assigned by MITRE, not by Rails; its generator is named as "CVE-Request-form 0.0.1", and the record does not say who made the request. It was reserved at 14:12:17.191 and published at 14:12:17.584, under a second apart (derived).

Dates from the Rails release records, the two CVE.org records and the GitHub advisory record, as read on 15 November 2026. Spans marked derived are this site's arithmetic.

Dated steps in the saved records, in UTC. Sources: the Rails release records, the CVE.org records, the NVD record, the GitHub advisory record and the DigitalOcean pull request, as read on 15 Nov 2026.

When (UTC)What the record saysWhere
4 Nov, 02:31:54Release 8.0.5.2 publishedRails release record
4 Nov, 02:46:00Release 8.1.4.1 published (14 min 6 s later, derived)Rails release record
4 Nov, 14:12:17CVE-2026-122006 published, MITRE, CVSS 3.1 9.8 (11 h 26 min after 8.1.4.1, derived)CVE.org record
4 Nov, 15:17:42NVD publishes its copy (1 h 5 min after the CVE record, derived)NVD record
4 Nov, 15:39:34A DigitalOcean pull request cites the CVE id and the advisory id (1 h 27 min after the CVE record, derived)Pull request record
5 Nov, 16:41:32CVE Program container adds three referencesCVE.org record
9 Nov, 21:12:06CISA coordinator stamp: Exploitation noneCVE.org record
10 Nov, 17:07:43CVE-2026-126082 reserved by GitHub_M (6 d 2 h 55 min after the first id, derived)Second CVE.org record
10 Nov, 21:08:05CVE-2026-126082 published (4 h 0 min after reservation, derived)Second CVE.org record
12 Nov, 16:02:27CVE-2026-126082 rejected as "a duplicate of another CVE", replacedBy CVE-2026-122006Second CVE.org record
12 Nov, 23:46:51Rails advisory published (7 h 44 min after the rejection, derived)GitHub advisory record

What the advisory record carries. Its identifiers list only CVE-2026-122006, yet its twelve references include the NVD page of CVE-2026-126082, the rejected id, as the saved record stood on 15 November. The record's updated time is one second after its published time, so the saved record shows no later update. The record also carries a field named as the NVD publication time, 10 November 21:19:26 UTC, while NVD's own record for CVE-2026-122006 says published 4 November 15:17:42. The 10 November time falls 11 minutes 21 seconds after the second id was published (derived). The saved pages do not say which record that field describes.

What the second id's record carries. Its rejection notice, the replacement pointer and the dates. It has no description and no references. Why GitHub reserved a second id on 10 November, six days after MITRE's record, and why it was rejected two days later: not stated. NVD's copy of the first record read "Awaiting Analysis" at 10:00 UTC on 15 November, and NVD shows no score of its own, only the one copied from MITRE's container.

Who is affected: three conditions against one sentence

The advisory's heading on this is "Affected applications": an application is affected only if it meets all of three requirements. The CVE record's description is one sentence: "the activestorage gem before 8.1.4.1" "when ImageMagick attachment processing occurs" can allow server read and write operations, because "there is not an ImageMagick security policy to block unintended functionality". So the description covers the first condition in outline and the second in one clause, and says nothing of uploads. The vector reads attack vector network, complexity low, privileges none, user interaction none. How the three conditions were weighed in the complexity rating is not stated.

Drawn from the Rails advisory, the CVE.org record, the Rails release notes and a third party report, as read on 15 November 2026. Amber cells differ or are partial, red cells are silent.

Who each saved page says is affected. Sources: the Rails release notes of 4 Nov, the Rails advisory of 12 Nov, the CVE.org record and the third party report, as read on 15 Nov 2026.

PageWhat it saysWhat it leaves out
Rails release notes, 4 NovWho must act: mini_magick below 5.1, PostScript or PDF rendering, setups that never receive the policy. "Every other application needs no change"Any list of what makes an app exposed; uploads; Debian and Ubuntu.
Rails advisory, 12 NovAll three: ImageMagick through the mini_magick processor; a host policy that does not deny the coders; "Allows file uploads from untrusted users"How many apps meet all three.
CVE.org recordImageMagick attachment processing; no security policy; vector AV:N, AC:L, PR:N, UI:NThe processor setting; uploads; the host policy's role.
Third party report (Turing repository)The same three requirements, plus "Vips is default for newer Rails"Any source for the Vips claim; the Rails advisory does not say it.

The processor default. The advisory says the mini_magick setting is "the default for applications on load_defaults 6.1 or earlier". It says nothing about what newer defaults are. The third party report says Vips is the default for newer Rails; the Rails pages saved here do not say that, and this briefing does not rely on it. The advisory ties the default to the load_defaults value, not to the gem version, so an application on a current Rails with an old load_defaults line falls under the first condition (this site's inference from the wording).

The host policy. The second condition reads "The policies that Debian and Ubuntu package do not", meaning those policies do not already deny the coders involved. The advisory does not name any other platform; whether a given host's policy denies them is for the operator to read.

Versions and scores as each saved page prints them. Sources: the Rails advisory (prose and structured fields), the CVE.org record and the third party report, as read on 15 Nov 2026.

PageAffected versions as printedScore and vector
Rails advisory, proseactivestorage < 8.0.5.2; activestorage >= 8.1, < 8.1.4.1Not in the prose
Rails advisory, structured fields< 8.0.5.2 (first patched 8.0.5.2); >= 8.1.0.beta1, < 8.1.4.1 (first patched 8.1.4.1)CVSS 3.0 vector AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, 9.8; CWE-1188; scorer not stated
CVE.org record, description"the activestorage gem before 8.1.4.1"CVSS 3.1, same letters, 9.8, in MITRE's container; CWE-1188
CVE.org record, structuredVersion 0 up to below 8.0.5.2; version 8.1 up to below 8.1.4.1As above
Third party report< 8.0.5.2 and >= 8.1, < 8.1.4.19.8, no vector

Read literally, the CVE description's words cover every version before 8.1.4.1, which includes 8.0.5.2, a release both the advisory and the CVE record's own structured data name as fixed. The structured data is tighter than the sentence (inference: the sentence is loose). The advisory's own prose starts the 8.1 line at 8.1 and its structured fields at 8.1.0.beta1; whether a prerelease is treated differently is not stated. The first range, below 8.0.5.2, has no lower bound, yet the fixed releases named are on the 8.0 and 8.1 lines only. Whether the maintainers issue a fix for older lines is not stated in the saved pages.

Third party material: a tiering report and a pull request

The Turing repository report. The CVE record's CVE Program container, dated 5 November, lists a report in the alan-turing-institute cyber-threat-observatory repository at a path containing 2026-11-04. That path is the only date: the saved text prints none of its own. Every claim below belongs to that repository, which is not the vendor. It says the flaw is "not currently exploited in the wild", assigns "TIER 2", states "wild_exploited: false" and "No CISA/Vulncheck KEV listing", and says "Vips is default for newer Rails". Its sections on exploitation mechanics and detection go further than the Rails text and are not reproduced here; the advisory says one chain was reported and describes none.

A difference in dates. The report is dated 4 November by its path, yet its text refers to a vendor advisory that notes a reported attack chain, to Debian and Ubuntu defaults, and to a workaround file path. The release notes of 4 November carry the mini_magick 5.1 and policy points but not the reported chain or the Debian and Ubuntu line, and the advisory record is stamped 12 November. The saved pages do not say whether the report was edited after 4 November, whether the advisory's text was readable earlier, or something else. This briefing uses the report only as a second description of the same bug.

The DigitalOcean pull request. The CVE record cites pull request 488 in DigitalOcean's droplet-1-clicks repository. Its title is "rails-24-04: bump Rails to 8.1.4.1 (Critical CVE-2026-122006 / GHSA-f3w6-w445-fwh9)". It was opened on 4 November at 15:39:34 UTC and merged on 10 November at 10:22:40 UTC (5 days 18 hours 43 minutes later, derived). It changes one line in one file, a version number in a template for a one-click Ruby on Rails image, from 8.1.4 to 8.1.4.1. The body carries the markers of a Cursor agent and describes the image's lockfile as listing the image_processing and mini_magick gems; it also has a "Not yet verified" heading for a newer image. A gem in a lockfile is not the variant processor setting, and the pull request does not say what that image's setting is. It shows that a downstream project cited both the CVE id and the 9.8 on the day of the record. This briefing claims nothing beyond that.

A third reference, a pull request to an advisory database, was listed by the CVE record and by the advisory and was not read. The advisory credits one HackerOne researcher as the reporter; this briefing does not print the handle.

What the saved pages say about exploitation

What each saved page prints about exploitation of CVE-2026-122006. Sources: the Rails advisory, the CVE.org record with its CISA container, the NVD record, the third party report and this site's KEV check, as read on 15 Nov 2026.

PageWhat it printsWhat it leaves out
Rails advisory, 12 Nov"One specific attack chain has been reported to us, but we do not assume it is the only one that exists."What the chain is; whether anyone used it.
CISA container in the CVE record, stamp 9 Nov 21:12:06 UTCSSVC: Exploitation none, Automatable yes, Technical Impact total, role CISA CoordinatorThe evidence; anything after 9 Nov.
Third party report, dated 4 Nov by its pathNot exploited in the wild, as that repository judgedIts date of judgement; any evidence.
Exploit likelihood field in the advisory recordEPSS 0.00741, percentile 0.54826The date of the figure.
CISA KEV, this site's check, 15 Nov between 09:20 and 09:30 UTCNo entryAnything about later.
NVD recordStatus Awaiting Analysis; copies the SSVC valuesAn analysis of its own.

None of this is a statement that the flaw was or was not used. The advisory says a chain has been reported to the maintainers; the CISA stamp and the third party report say no exploitation was known at their dates; no saved page reports an attack, a victim or a count. All of it can be overtaken: the 9 November stamp is 5 days 12 hours old at the 15 November reads (derived), and the KEV check is a single moment.

Labels that do the work of a control

9.8 and CRITICAL. A vector has fixed letters and no field for the three conditions; a score of 9.8 describes the worst case in which all three hold, not the share of Rails applications that are exposed. Nothing saved states that share.

"Possible". The advisory's title says "Possible arbitrary file write and read in image analysis", its impact text says an application "may allow" an unauthenticated attacker to write files, and its details say an uploader of a crafted file "may be able to invoke one". The wording is conditional; the severity label is not.

"Default". The weakness class is CWE-1188, insecure default. The advisory's own conditions put the default on the processor setting and the exposure on a host policy and on who can upload; a default on one line is not a control on the others.

A distribution's packaged policy. The advisory says the Debian and Ubuntu policies do not deny the coders involved. A host that runs a packaged policy has not thereby met the second condition's safe side.

"Needs no change". The notes say every other application needs no change, after a list of who must act. The list is the test, not the sentence.

The policy, the setting and the variable, as the advisory prints them

These strings are copied from the Rails advisory of 12 November 2026, read on 15 November 2026, and the mechanism also appears in the release notes. This site ran none of them and adds nothing to them. The saved pages print no indicator, log line or detection rule for abuse and no description of the reported chain, so none appears here.

The workaround policy. For those who cannot upgrade at once, the advisory says to copy the rules of the policy that the fixed versions ship into ImageMagick's system policy, and that at a minimum the application must deny the MSL coder, the Ghostscript coders and the gs delegate. The three lines it prints are below. ImageMagick applies every policy file on its search path, so these add to whatever the host already has.

Text
<policy domain="coder" rights="none" pattern="MSL"/>
<policy domain="delegate" rights="none" pattern="gs"/>
<policy domain="coder" rights="none" pattern="{PS,PS2,PS3,EPS,EPS2,EPS3,EPI,EPSF,EPSI,EPT,EPT2,EPT3,PDF,AI,EPDF,PDFA,POCKETMOD,XPS,PCL}"/>
The three policy lines as printed in the Rails advisory of 12 November 2026, read 15 November 2026.

Where the advisory says to put them. The advisory gives this path as an example, with the words "for example": the system policy of what the directory name suggests is an ImageMagick 7 installation (this site's reading). Where your host keeps its policy may differ.

Text
/etc/ImageMagick-7/policy.xml
The example system policy path as printed in the Rails advisory, read 15 November 2026.

The setting that puts an application under the first condition. The advisory names it, and says it is the default on load_defaults 6.1 or earlier. Search your configuration for it, and for your load_defaults line.

Text
config.active_storage.variant_processor = :mini_magick
The processor setting as printed in the Rails advisory, read 15 November 2026.

How the fix reaches ImageMagick. The fixed versions prepend the policy directory to an environment variable through a MiniMagick hook. This is why an application that replaces the hook, clears the environment or installs its own shell does not receive the policy, as the diagram shows.

Text
MAGICK_CONFIGURE_PATH through MiniMagick.cli_env
The mechanism as worded in the Rails advisory, read 15 November 2026.

The version the policy needs. Applying the policy requires this minimum mini_magick version, and the advisory says an older one makes the application raise an exception during boot.

Text
mini_magick >= 5.1
The minimum version as printed in the Rails advisory, read 15 November 2026.

Where the sources differ

Differences between the saved pages and how this briefing handles each. Sources: all eight saved pages, read 15 Nov 2026.

TopicThe differenceHandled as
Version range wordingCVE sentence: before 8.1.4.1. Advisory and CVE data: below 8.0.5.2, and the 8.1 line below 8.1.4.1. Advisory prose starts 8.1 at 8.1, its structured field at 8.1.0.beta1.Shown side by side; the data read as tighter (inference).
CVSS versionAdvisory field CVSS 3.0, CVE record 3.1; same metrics and score, 9.8.Both reported; scorer of the 3.0 vector not stated.
Who is affectedCVE sentence names ImageMagick and no policy; advisory lists three conditions.Both quoted; neither resolved.
Processor defaultAdvisory: the default on load_defaults 6.1 or earlier. Third party: Vips is default for newer Rails.The advisory followed; the third party claim attributed.
Report date and contentThird party report dated 4 Nov by its path; its content matches the advisory of 12 Nov.Reported, not explained.
The second idRejected on 12 Nov; the advisory of 12 Nov still lists it among references.Reported, cause left open.
Advisory record's NVD fieldField says 10 Nov 21:19:26 UTC; NVD says 4 Nov 15:17:42.Both reported; the field's subject not stated.

What the pages state, and what they do not

What is stated on a saved page and what is not, topic by topic. Sources: all eight saved pages, read 15 Nov 2026.

TopicStated on a saved pageNot stated
The fixRelease 4 Nov; the changelog names the advisory id; policy shipped and prependedWhether it breaks PostScript or PDF apps beyond the re-allow step.
The CVE recordMITRE, 4 Nov 14:12:17 UTC, 9.8, CWE-1188Why it existed before the advisory; who requested it.
The second idReserved 10 Nov, published 10 Nov, rejected 12 Nov as a duplicate, replaced by the firstWhy it was reserved, why it was rejected, why the advisory lists it.
The advisoryPublished 12 Nov 23:46:51 UTC; three conditions; one reported chainWhy 12 Nov; what the chain is.
ExposureThree conditions that must all holdHow many apps use mini_magick, a permissive policy and untrusted uploads.
ExploitationNone known at the CISA stamp and the third party report; a chain reported to the maintainersAny attack, victim or count; anything after those dates.
Older Rails linesThe range has no lower boundWhether any fix exists below 8.0.
The reporterA HackerOne researcher is creditedAnything else; no handle is printed here.

What to do, in order

Take this with you

The vendor's steps first, then this site's suggestions

  • The advisory's step: upgrade activestorage to a fixed version, 8.0.5.2 on the 8.0 line or 8.1.4.1 on the 8.1 line. For those who cannot upgrade at once, the advisory publishes two patches, one for 8.1 and one for 8.0.
  • The advisory's step: an application whose variant processor is mini_magick must use mini_magick 5.1 or later, which applying the policy requires; with an older version the application raises an exception during boot.
  • The advisory's step: an application that renders PostScript or PDF through ImageMagick must re-allow the gs delegate and the coders it uses in the host policy file.
  • The advisory's step: an application that replaces MiniMagick.cli_env after boot, clears the environment with MiniMagick.cli_prefix or installs its own MiniMagick::Shell does not receive the shipped policy and must carry its rules in ImageMagick's system policy, as must one using GraphicsMagick.
  • The advisory's step: if attachments are on the Disk service, change secret_key_base and every secret the application process can read, including the master key, everything the credentials file holds, storage and database credentials and third party tokens. Do not keep an exposed secret as a fallback.
  • This site's suggestion, to do first alongside the upgrade: find out whether your application's Active Storage variant processor is mini_magick, and what your load_defaults line is. The advisory's conditions name mini_magick.
  • This site's suggestion: read your host's ImageMagick policy and check whether it denies the coders the advisory lists, rather than assuming a packaged policy does.
  • This site's suggestion: the fix has been available since 4 November 2026 (11 days 7 hours before the 10:00 UTC read of 15 November, derived), so measure the exposure window: which applications let untrusted users upload, and since when each ran an affected version.
  • This site's suggestion: whether to rotate secrets without any evidence of abuse is a risk decision, not the advisory's instruction. The advisory's rotation step is written for applications on the Disk service.
  • This site's suggestion: if you run Rails below 8.0, read the advisory's affected range yourself and decide on the system policy workaround; the saved pages name no fixed release on those lines.

The question the records leave open

A 9.8 record sat on CVE.org for 8 days 9 hours before the Rails advisory listed the conditions that decide who is exposed, so who will say why the order ran that way, and how many applications met all three conditions in the meantime?

Key facts

Sources

  1. PrimaryAdvisory GHSA-f3w6-w445-fwh9, Active Storage: Possible arbitrary file write and read in image analysis. Published 12 November 2026 23:46:51 UTC: impact, affected applications, mitigation, workaround policy, versions, references. Saved by the lead on 15 November 2026GitHub, the Rails maintainers' advisory recordaccessed 2026-11-15
  2. PrimaryRelease 8.0.5.2, published 4 November 2026 02:31:54 UTC, with the full changelog and the Active Storage ImageMagick policy item. Saved by the lead on 15 November 2026Rails (GitHub release record)accessed 2026-11-15
  3. PrimaryRelease 8.1.4.1, published 4 November 2026 02:46:00 UTC, with the full changelog and the same Active Storage item. Saved by the lead on 15 November 2026Rails (GitHub release record)accessed 2026-11-15
  4. PrimaryCVE record CVE-2026-122006: assigner MITRE, published 4 November 2026 14:12:17 UTC, description, CVSS 3.1 9.8, CWE-1188, the CVE Program container of 5 November and the CISA container of 9 November. Saved by the lead on 15 November 2026CVE Program (cveawg.mitre.org)accessed 2026-11-15
  5. PrimaryCVE record CVE-2026-126082: assigner GitHub_M, reserved 10 November 2026 17:07 UTC, published 21:08 UTC, rejected 12 November 16:02 UTC as a duplicate, replaced by CVE-2026-122006. Saved by the lead on 15 November 2026CVE Program (cveawg.mitre.org)accessed 2026-11-15
  6. PrimaryNVD record CVE-2026-122006 (API 2.0), status Awaiting Analysis, last modified 12 November 2026 22:09 UTC, served 15 November 10:00:19; copies the MITRE vector as a Secondary metric and the CISA SSVC values. Saved by the lead on 15 November 2026NIST National Vulnerability Databaseaccessed 2026-11-15
  7. Reported byTiering report for CVE-2026-122006, dated 4 November 2026 by its path only: not exploited in the wild, tier 2, says Vips is default for newer Rails. A third party, not the vendor; used only as a second description of the same bug. Saved by the lead on 15 November 2026Alan Turing Institute cyber-threat-observatory repository (third party)accessed 2026-11-15
  8. Reported byPull request 488, a one line version bump in a one-click Ruby on Rails template citing CVE-2026-122006 and GHSA-f3w6-w445-fwh9; opened 4 November 2026 15:39 UTC, merged 10 November 10:22 UTC. Cited by the CVE record. Saved by the lead on 15 November 2026DigitalOcean droplet-1-clicks repository (GitHub pull request record)accessed 2026-11-15

Share this briefing

Know someone who owns this problem? Send it to them.

Related briefings

The briefing, in your inbox

Practitioner analysis of cyber and AI security news. No vendor noise.

How often

Every new briefing in one email, at 7am, or at 7am, 12:30pm and 6pm. Nothing is sent when nothing is new. Unsubscribe any time.