Repository navigation
Releases
CookieRadar 2026.48
Changed
-
Two OpenSSL issues of 2026-11-17 are accepted until Debian ships 3.5.10. On 2026-11-18 Docker
Scout reported CVE-2026-107493 and CVE-2026-124709 as high against the same openssl, and Security
Posture failed again. OpenSSL rates both Low and fixes them in 3.5.10; Debian lists trixie and sid
as vulnerable, with no fix. Neither is reachable here: one needs an application passing its own
stack toPEM_X509_INFO_read_bio_ex(), which OpenSSL and Python's ssl never do, the other a QUIC
server. The advisory's third issue, CVE-2026-35190, is not reported by any scanner yet and has no
entry, since an unused one fails the gate. Reviewed on 2026-12-15 with the four of 2026-11-10 below. -
Four OpenSSL issues of 2026-11-10 are accepted until Debian ships 3.5.10. At the first scan of
main on 2026-11-17 Docker Scout reported CVE-2026-95691, CVE-2026-95692, CVE-2026-88823 and
CVE-2026-107496 as high against the base image's openssl 3.5.7-1~deb13u3, and Security Posture
failed. OpenSSL rates all four Low and fixes them in 3.5.10; Debian has no trixie package with the
fix yet. None is reachable here: they need the OpenSSL QUIC client, AES-GCM-SIV through EVP, or
CMS_encrypt(), and Python's ssl exposes none of them, while the audited sites go through
Chromium's own BoringSSL. The entries are reviewed on 2026-12-15; the rebuild that brings the DSA
will make them fail the gate as unused. -
The 49 exceptions due on 2026-11-30 are reviewed. Without the review, Security Posture would
have failed on its own from 2026-12-01. Each entry was read again on 2026-11-17 against the Debian
security tracker and archive, and against the image. 37 move on. libxml2, expat 2.9.0, krb5 and
cups are still unfixed in trixie, or fixed only in sid, and move to 2026-12-31; acl, which is
no-dsa and waits for a point release, moves to 2027-01-31. 12 go. Two expat CVEs were fixed by
DSA-6566-1 and the image has the fix. harfbuzz is not affected in Debian. Four gnutls CVEs are
rated unimportant by Debian, and one was rejected by its CNA. The four xorg-server entries cover
packages the Dockerfile purges. krb5 CVE-2026-107778 stays: Docker Scout has been quiet on it
since 2026-10-11, but the package has not changed. Run against the live alerts, the gate is green
on 2026-11-17 and on 2026-12-01.
Fixed
-
A citation is "verified" only when the reference it cites is in the text downloaded now. The
check looked at the article: with Cellar's page parsed and a point it cites, such as GDPR art.
5(1)(a), missed by the parser, the reference fell back to the cache, old SHA-256 and date included,
under "verified against Cellar"; with an empty cache the same label stood besideSHA256: not available. Every cited reference is now looked up in the text just parsed, and one that is missing
makes the act a fetch failure: its citations come from the cache, labelled as such, or have no hash
and say so. Ported from mailradar #103, as the five law modules below. -
A failed download always reaches the EUR-Lex fallback and the cache. Only
httpx.HTTPErrorwas
a fetch failure. A 303 tohttp://xn--/DOC_1makes httpx raiseidna.core.IDNAError, a
ValueError, which skipped the fallback and ended the whole law check.httpx.InvalidURLand
ValueErrorare now fetch failures too. -
Redirects stay on the source's host. Any
Locationwas followed, so a text served by another
host was labelled "verified against Cellar". Cellar may redirect only to publications.europa.eu,
EUR-Lex only to eur-lex.europa.eu and Normattiva only to www.normattiva.it, on the default port; the
http:// hop Cellar makes is still taken over HTTPS. The hosts indocs/nis2-compliance.mdare now
the only ones the law check can reach, and the document says so. -
A page is read up to 32 MiB, and a fetch ends after 120 seconds. The timeout bounded each read,
not the whole, and nothing limited the size of the page. The body is now streamed, refused past
MAX_RESPONSE_BYTES(32 MiB; the GDPR from Cellar is 823981 bytes) and abandoned past
FETCH_DEADLINE(120 s, redirects included); either one is a fetch failure, so the fallbacks apply. -
A change in the law is reported even when the cache cannot be written. The comparison with the
previous text was made by the save, so with a read-only cache a law whose text changed was cited
without the warning, and an unchanged wording was dated to the day of the audit. The merge is now
law_cache.merge, made before the save and kept whether the save succeeds or not. -
Errors and status lines name the source that was tried. A missing article always read "not
found in the EUR-Lex page", for Normattiva and Cellar pages too, and the report's status line said
"EUR-Lex unreachable" whatever had happened — for the GDPR, whose first source is Cellar, even when
Cellar had answered.LawFetchErrorcarriessourceandreached,ActStatuscarriestried
andfailure, and the line says "GDPR: Cellar answered without the text cited" or "GDPR: Cellar
and EUR-Lex unreachable". -
A corrupt law cache is an empty one. A deeply nested
law_cache.jsonmadejson.loadsraise
RecursionError, and the whole law check failed. It is now read as an empty cache. -
Normattiva's "-bis" paragraphs have a reference of their own. Paragraph numbers were read as
\d+, so paragraph 2-bis of art. 122 of the Codice privacy, and 3-bis and 4-bis of art. 21 of the
Consumer Code, were part of the article's text and could not be cited; the later Latin numbers
(-ter, -quater, ... -decies, -terdecies) are read too. -
auditno longer lists trackers "loaded after rejection" when nothing was rejected. With a
tracking cookie stored before any choice the verdict is VIOLATION whether or not the reject button
was found, and the report then printed "N tracker(s) loaded after rejection" for the trackers of a
session in which the refusal was never applied: on www.repubblica.it on 2026-11-17, a pay-to-refuse
wall, five of them. The list is now printed only when the refusal was clicked, asbatchand the
law check already did. -
A site that never loaded is no longer reported as measured.
verdictignored a session's error
when there was no status: a navigation that failed outright (name not resolved, connection refused
or reset, nothing before the timeout) left a session with no trackers and no cookie, which read as a
clean one. On 2026-11-17 www.gnu.org's pre-consent session gotnet::ERR_CONNECTION_RESETwhile the
other two loaded, and the verdict was OK; an unresolvable host exited with 2 and "reject button not
found". A session with an error, no status and nothing from a document is now unreachable
(scanner.unreachable). Every session unreachable is⛔ NOT MEASURED, exit 3, inaudit,batch
and the law check; session 1 or 3 unreachable is never OK, and UNVERIFIED says which session never
loaded. The README's exit codes say so. -
A close "×" that subscribes is no longer clicked as a refusal. The close step ran before the
check for a refusal sold with a subscription, and chose by selector alone. On www.ansa.it on
2026-11-17 iubenda's.iubenda-cs-close-btnwas labelled "Rifiuta e abbonati": it was clicked,
Piano's checkout opened, and the session was measured as a refusal and cited. Every step that
clicks now skips a control whose aria-label, text, value or title is a refusal sold with a
subscription, and the English "Reject all and subscribe" (also "Reject", "Decline" or "Refuse" …
"and subscribe") is recognised beside "Rifiuta e abbonati". -
A tracker fired right after "Reject" is reported. The post-reject session waited two seconds
after the click and only then emptied its lists before the reload, so every request in those two
seconds was thrown away — among them a tag that fires on the consent update whatever the choice,
which is the misconfiguration the session exists to find. The lists are now cut where they stood
when the refusal was clicked: what the page loaded before it still does not count, what the click
set off does. -
A tag manager's library and a Consent Mode ping with storage denied are no longer cited after a
refusal. Every tracker host contacted after "Reject" was a violation of GDPR art. 5(1)(a),
ePrivacy art. 5(3), art. 7 and the Consumer Code, including the two kinds of request the README
already called possibly lawful before any choice. On 2026-11-17, after a refusal that was clicked,
www.köln.de loadedgtm.js,gtag/jsand a GA4 collect withgcs=G100, and nothing else; it was
a VIOLATION with four provisions cited. www.comune.milano.it loaded nine scripts from
assets.adobedtm.comand no Adobe beacon. A domain whose every request after the refusal was a tag
manager's library fetched as a script (Google Tag Manager, Adobe Launch, Tealium, TagCommander) or
a Google ping withgcs=G100is now reported as "Loaded after rejection, not cited", with the
reason, and is not cited — unless the session holds a tracking cookie by name or a cookie on that
domain. Beacons, pixels and granted pings stay violations, and so does a domain with any other
request: aranzulla.it's doubleclick.net and googlesyndication.com still are.Violationshas a
not_citedfield;scanner.not_cited_after_rejectionnames them. The README says what the rule is. -
A listed site's own cookies are not "tracking stored before any choice". A cookie counted as
tracking storage when its domain was on the tracker list, and linkedin.com is on it as a tracker of
other sites: audited on its own domain, whose requests were already left out as first-party, its
bcookieandlangmade a VIOLATION citing art. 122(1) of the Codice privacy. A cookie now counts
by its domain only when that domain is not the site's (is_same_sitewith the ...
CookieRadar 2026.47.1
Changed
- A change to the image earns a fix version, and only that. A version ahead of the last tag needed a
change undercookieradar/, so that no Radar gets a number for nothing. The Playwright fix above is
in the Dockerfile alone, and the rule refused the release that puts a fixed image on Docker Hub, where
latestis what Docker Scout reads and what people pull. A change to the Dockerfile, what the image
is built from, now earns the third segment, the one kept for something urgent on what has already
shipped, and never an ordinary count. The package on PyPI is the same code as 2026.47.
tests/test_version_contract.py.
Security
- Playwright 1.65.1, for the Node it carries. Playwright's driver is a Node binary of its own, and
on 2026-11-15 Docker Scout reported ten high CVEs against the Node 24.21.0 inside 1.63.0, in the image
just published as 2026.47: three fixed in Node 24.22.1 (CVE-2026-105482, -105483, -106069) and seven
in the security release of 2026-11-11, 24.22.3 and 26.13.1 (CVE-2026-123863, -123864, -123865,
-123866, -123872, -123874, -126051). Read from the Linux wheels that day, 1.64.1 bundles Node 24.22.1
and 1.65.0 bundles 26.13.0, released before the fixes; 1.65.1 is the first with 26.13.1. It is four
days old, under the fourteen days a dependency waits here, and goes in because the version it replaces
has the known vulnerabilities. Its Node 26 links libatomic.so.1, which the slim base lacks and which
stopped the build on 2026-11-12, so the Dockerfile installs libatomic1 beforeplaywright install-deps. pyee leaves the lock: Playwright no longer depends on it.
tests/test_docker_playwright.pyholds the floor and the order.
CookieRadar 2026.47
Added
scanner.verdictandscanner.Verdict: the verdict as a public function. It lived in
cli._verdict, which a caller could not import, and APKRadar's--fulldecided it by itself,
more narrowly: three of the four ways a site is not OK came out as compliant. The command line
now uses the same function, andVerdict's values are its exit codes.
Changed
-
The image runs Python 3.14. It was built on python:3.12-slim-trixie: 3.12 has had security fixes
only since April 2025, and 3.13 joined it in October 2026. 3.14 gets bug fixes until October 2027 and
security fixes until October 2030; 3.15, a month old, was too new for the wheels the image needs.
exeradar and patchradar were already built on 3.14.sbom.ymlresolves the Python environment of a
release with the image's interpreter, so it moves to 3.14 too, andtests/test_docker_python.pykeeps
the two together. Measured on 2026-11-13 on the rebuilt image: the same audit of a local page, on 3.12
and on 3.14, gives the same verdict, provisions and report. -
pip leaves the image once cookieradar is installed. The 3.14 base ships pip 26.2.2, and Trivy
reported five high findings against the packages pip vendors: msgpack 1.1.2 (CVE-2026-114613,
GHSA-6v7p-g79w-8964), urllib3 2.7.0 (CVE-2026-97687, -97689) and setuptools 70.3.0 (CVE-2025-47273).
pip 26.3 still vendors that setuptools. Nothing in the image runs pip after the build, so the layer
that installs cookieradar uninstalls it, andsbom.ymldrops it from the venv the source SBOM
describes, astests/test_nis2_evidence.pyalready required of any package the image removes
(tests/test_docker_pip.py). -
Trivy's expat and libXi findings are recorded, measured in the rebuilt image. On 2026-11-09
Trivy began reporting four expat CVEs (CVE-2026-115506, -113351, -118370, -117610) and two libXi
ones (CVE-2026-109089, -109090) in the Chromium dependencies, and Security Posture failed. Measured
that day in the image rebuilt from v2026.46: Chromium declares the system libexpat directly, which
the entries had not said (they named fontconfig and Mesa alone), so the expat entries rest on what
each flaw takes rather than on its reach, and the two Debian has fixed elsewhere keep the short
date, for the stable update to ask for. libXi comes in through the accessibility bridge only and
its flaws need an X server, which headless Chromium never connects to. -
The sqlite3 findings say what Debian decided, and Trivy's two new ones are recorded. On
2026-11-09 Trivy began reporting CVE-2026-119374 and CVE-2026-119379, which only Docker Scout had
reported, and Security Posture failed on main. Both now have a Trivy entry beside the Scout one.
The sqlite3 entries also said there was no fixed version anywhere and that they waited for Debian:
Debian has since fixed the three CVEs in sid (3.54.1-1) and marked them no-dsa for trixie, so no
trixie update will come. They stay accepted on the ground they already gave, that no code of the
project reaches the flawed parts of SQLite, until 2026-12-31, when the question becomes a base
image with a fixed sqlite3. -
Dependabot proposes a release once it is 14 days old, and actions arrive as one pull request.
The Radar take a dependency at least 14 days after its release unless a known vulnerability makes the
older version the risk; without a cooldown Dependabot opened twelve pull requests across the Radar on
2026-10-29 and 2026-11-02 for releases two to thirteen days old..github/dependabot.ymlnow sets a
14-day cooldown on every ecosystem — Dependabot's security updates are not subject to it — and groups
the GitHub Actions updates into one weekly pull request.tests/test_dependabot_config.pyholds both. -
Trivy scans the image a pull request builds; Snyk is gone. As in maksimtech/apkradar#54: of what
snyk.ymlran, only the container scan worked. Snyk Open Source stopped onSNYK-OS-PYTHON-0013
("Missing required packages") here too, so its check on new dependency issues let everything through;
Snyk Code repeated CodeQL, Bandit and SonarCloud; and the free plan's monthly test quota, shared by the
Radar, was spent.trivy.ymlnow builds the image and scans it before merge, with no token and no
quota. The Snyk-sourced entries leftSECURITY-EXCEPTIONS.toml, their reasons moving into the entries
of the scanners that still report the same CVEs;tests/test_trivy_workflow.pyholds the image scan,
andtests/test_workflow_set.pykeepssnyk.ymlretired.
Its first run reported the nine base-image CVEs the other Radar record (util-linux ×4, ncurses,
systemd, perl, sqlite3, acl), to wait for Debian, and seventeen in Chromium's and the virtual
display's dependencies: libxml2 ×8 (CVE-2026-6653 critical), cups ×2, the X.Org server behind Xvfb ×5,
libX11 and libXrender. Each entry says, measured in the published image, why it is not reached: Chromium
does not link libxml2, the image has the cups client library and no print server, Xvfb is never
started and headless Chromium talks to no X server. The fix-elsewhere ones keep a short review date
and the Debian bug to ask on. The Snyk entries' reasons moved to the Docker Scout and Trivy entries of
the same CVEs; expat CVE-2026-115506, which only Snyk reported, is noted in the header. With them, the
three sqlite3 CVEs Docker Scout raised onmainon 2026-11-02, which had turned Security Posture red. -
The author address published on PyPI is the project alias, github@maksimtech.com. It was a
personal mailbox, which PyPI shows to anyone and bulk senders harvest: on 2026-10-31 a cold "patch
offer" from a mass-mailing account reached it. Releases already published keep the old address;
the next one carries the alias.tests/test_package_metadata.pykeeps it that way.
Fixed
-
Security Posture reads every page of alerts. The code scanning API returns at most a hundred alerts
per call, and the gate read one page of each list. On 2026-11-13 #68 had 123 open alerts on its ref and
main had 142: three exceptions whose findings sat past the hundredth read as matching nothing, and the
gate failed on a product whose every alert was explained. Read the other way, an unexplained alert past
the hundredth would not have been seen at all. Each of the four reads now goes through
gh api --paginate --slurpand joins the pages; a failed read still fails, so an unreadable list is not
taken for an empty one.tests/test_security_posture_workflow.pyruns the step's own function against a
stand-inghthat answers in two pages. Run against the real API on #68: 123 alerts read, verdict ok. -
The image builds again: it installs the Playwright that uv.lock resolves. The Dockerfile installed
playwright>=1.40.0, so each build took the newest release on PyPI. On 2026-11-12 that was 1.65.1,
published the day before, whose Node driver needslibatomic.so.1: python:3.12-slim-trixie does not
ship it,playwright install-deps chromiumstopped with exit code 127, and Trivy and Security Posture
failed on main with no image to scan. The build of 2026-11-09, on the same base image digest, had
taken 1.64.1. The image now installsplaywright==1.63.0, the version uv.lock resolves, and
tests/test_docker_playwright.pyfails when a lock update leaves the Dockerfile behind.
docs/nis2-compliance.mdsaid the image resolves every dependency at build time; it now names
Playwright as the exception. -
Docker Scout no longer fails on its own base-image advice. From 2026-10-24 every run of
docker-scout.ymlfailed and GitHub mailed it on each push to main: the CVE scan worked, then the scout
CLI crashed in therecommendationspart (nil pointer inrecommendations/markdown.go) on images built
on python:3.12-slim-trixie. The scan is now a step of its own that writes the SARIF and fails the job if it
fails; the recommendations are a second step withcontinue-on-error.tests/test_scout_steps.py
holds the two apart. Same change as maksimtech/mailradar#76. -
Docker Scout scans the image as soon as it is published, and Security Posture reads that scan.
The scan readslatestfrom Docker Hub but ran only on a push to main, weekly or by hand: after the
certifi release the gate stayed red on the replaced image until both were dispatched by hand.
docker-scout.ymlnow follows the image workflow (workflow_run, successful runs only) and
security-posture.ymlfollows the scan.docs/nis2-compliance.mdnow says so, andtests/test_scan_after_publish.py
holds them together.
Security
-
The image no longer carries an X server.
playwright install-deps chromiuminstalls xvfb and
xserver-common for headed browsers without a display; nothing in the image starts one, and nothing
else installed depends on either (measured 2026-11-13). The first image scanned since the build broke
brought six X.Org Server CVEs against them (CVE-2026-109091, -109092, -109093, -109095, -109096,
-109903), fixed in sid and open in trixie. Both packages are now purged in the same layer, and only
they:--auto-removewould also take Mesa and libvulkan, which Chromium may load. The five older
X.Org findings accepted in SECURITY-EXCEPTIONS.toml close with them.--no-headlesshas no display
in the container.tests/test_docker_xserver.pykeeps both packages out. -
Fourteen MIT krb5 CVEs are recorded in SECURITY-EXCEPTIONS.toml. MIT's announcement of
2026-11-09 (fixed in krb5 1.22.3) has no fix in Debian, sid included (bug 1153598). Measured
2026-11-13: krb5 comes in through libcups, which chrome-headless-shell does not declare; it names
libgssapi_krb5 only to answer an HTTP Negotiate challenge, with no krb5.conf and no credential cache in
the image. Three of the CVEs are in libgssrpc, which is not installed (this line said two until the
Docker ...
CookieRadar 2026.46
Security
- certifi 2026.10.19 in
uv.lock, from 2026.7.22. The older bundle trusts emSign root CAs whose
intermediate CA keys were compromised (GHSA-6q48-6c7w-9gw2, critical, published 2026-10-20);
2026.10.19 is the release the advisory names as fixed. The Security Posture gate stopped every pull
request on it. - expat CVE-2026-115506 accepted in
SECURITY-EXCEPTIONS.toml(SNYK-DEBIAN13-EXPAT-21479164,
reported on 2026-10-23, review by 2026-11-30). A denial of service through entity amplification,
unfixed in Debian and upstream; in the image the system libexpat1 only reads fontconfig's and Mesa's
own files, and cookieradar parses no XML. - harfbuzz CVE-2026-113679 accepted in
SECURITY-EXCEPTIONS.toml(SNYK-DEBIAN13-HARFBUZZ-21663516,
reported on 2026-10-24, review by 2026-11-30). The flaw is in HarfBuzz's experimental WebAssembly
shaper, which Debian does not build: thewasmoption is disabled in 10.2.0 and libharfbuzz0b has no
WAMR runtime among its dependencies.
Fixed
- A verified text names the source it came from. EU acts are downloaded from Cellar
(publications.europa.eu), with EUR-Lex only as the fallback, but the report said "verified against
EUR-Lex" whichever answered (open item of the 2026-10-17 audit). It now says Cellar, EUR-Lex or
Normattiva, as the case was;law_checker.checktakes an optional HTTP client, so the path is tested
on recorded answers.law_fetcherstays the only module that imports httpx at run time
(test_nis2_evidencenow tells a type-checking import from a real one).
CookieRadar 2026.45
Added
docs/nis2-compliance.md: the answers to the seven questions a customer subject to NIS2
asks a software supplier, with only what the repository can back. Written in Italian, for
customers under d.lgs. 138/2024. Where the answer is "no" it says no: there is no penetration
test, no second reviewer, no access control because a CLI has no users, and the maintainer
can bypass the branch rules. It says what leaves the machine: the audited page and every
third party it loads, three times, from the operator's own address; the CELEX number of a
cited act to publications.europa.eu (eur-lex.europa.eu as the fallback) and the article of
the Consumer Code to Normattiva, only whenauditcites a provision. It says what Chromium
runs without (its sandbox, as Playwright starts it) and what was measured of its own traffic
on 2026-10-11: nothing but the page.tests/test_nis2_evidence.pycompares the hosts in the
document with those named in the modules that import httpx or Playwright, and holds the
cache, the report, the browser state, the image user, the release files and the scanners to
what the document says; it refuses "NIS2-ready", "no data sent to third parties" and similar
claims anywhere in it, inSECURITY.mdor in the README.- Every release carries SBOMs and dated scans (
.github/workflows/sbom.yml): CycloneDX 1.5
of the image, Debian packages included (Syft), and of the Python environment
(cyclonedx-py), plus Trivy reports of the image and of the repository at the tag, and a
note saying which commit and image digest they describe and that a CVE scan is not a
penetration test. It runs after "Publish to PyPI" has pushed the image of the tag, and on
demand for a release that is already out. SECURITY.mdsays how vulnerabilities are found, how fast they are answered, how
customers are told and how to roll back. A fix ships with the next release; for CVSS 9.0
or more, customers who gave a security contact hear within 48 hours. The paragraph on
"Known Open CVEs", last reviewed on 2026-09-10, is gone: it blamed Debian Bookworm, and the
image has been built on Trixie since 2026-09-20;SECURITY-EXCEPTIONS.tomlis where the accepted ones
are, each with a review date.uv.lock, written with uv 0.12.18, and a Quality step that fails when it falls behind
pyproject.toml.
Fixed
- Tracking stored before any choice never reached the verdict. The verdict read the session
after the refusal only, so a site that wrote Google Analytics'_gabefore its banner was
answered — repubblica.it on 2026-10-17, TCF string still empty — could end in exit 0. A tracking
cookie present at the end of the pre-consent session (_ga,_ga_<id>,_fbp,_gcl_au, or
any cookie of a known tracker's domain) is now a VIOLATION of its own, whatever the refusal did,
with the same exit code, and the report cites art. 122(1) of the Codice privacy, fetched from
Normattiva, with ePrivacy art. 5(3). A request to a tracker before any choice is listed and not
cited: it does not by itself show anything stored on the device.batchsays the cause too. - corriere.it's refusal with a subscription was not seen, because it is a link. On 2026-10-17
the wall (RCS's, in the page itself — not in an iframe, as the audit had recorded) offered
"accetta e continua" and "preferenze" as buttons and "Rifiuta e abbonati" as an<a href="#">.
The subscription label was looked for among buttons only, so the wall showed no refusal of any
kind. Links are searched too; the report names the paid refusal and cites nothing. - Cookiebot was not recognised, and the report said there was no banner. On cookiebot.com on
2026-10-17 the dialog offered "Deny", "Customize", "Allow all" (in English, the scanner's
locale; "Rifiuta" and "Accetta tutti" in Italian, with the same ids), and CookieRadar knew
neither the labels nor the ids, while its banner selectors are case-sensitive and missed
"CybotCookiebotDialog". Didomi's notice names none of the words the hint looks for. Cookiebot's
ids and the labels measured that day ("Deny", "Allow all", corriere.it's "Accetta e continua")
are known now, the hint ignores case and knows Didomi's prefix, and when it recognises nothing
the note says no banner was recognised instead of claiming none was there. - A banner shown only to a browser that is not headless was never seen. Headless Chromium
announces "HeadlessChrome/", and on subito.it on 2026-10-17 the Didomi notice was not
mounted for it and was for "Chrome/" from the same browser: the audit said
UNVERIFIED, "no refusal could be applied", about a site that offers one. The sessions now
announce the browser's own User-Agent as the build says it with a window, and the report
states it. The test runs a real Chromium against the notice captured that day. - GDPR art. 4(11) was cited against banners that offer a refusal. On aranzulla.it (iubenda)
the refusal is the banner's "×", whose text says "Chiudi questa informativa per continuare
senza accettare"; on repubblica.it it is "Rifiuta e abbonati". Both were reported on
2026-10-17 as "Accept was applied and no refusal control was found", with art. 4(11) and
ePrivacy art. 5(3) cited. The close "×" of iubenda, Didomi, OneTrust and Cookiebot is now
tried after every explicit refusal, visible only — repubblica.it keeps one in the page with
display: none. A refusal sold with a subscription is named in the report, not clicked, and
cites nothing. And "no refusal" now also needs Accept to have been the one control on the
banner, counted just before the click: with others beside it, none recognised, the note gives
their number and cites nothing. The tests run a real Chromium against the two banners as
captured that day, trimmed, ids and order kept. - The local HTTPS server of the law-fetch tests allowed any TLS version. CodeQL raised a
highpy/insecure-protocolalert on it, which the new pull-request gate caught on #37. It now
requires TLS 1.2 at least, and makes its certificate withcryptography, added to thetest
extra, instead of callingopenssl, so the test can no longer skip itself. - On a pull request, Security Posture called live exceptions unused. CodeQL reports only what
the change introduces on a pull request ref, so when it reported anything, an exception for a
finding still open on main in an untouched file matched nothing and failed the gate
(cookieradar #37, 2026-10-12). main's open alerts now count as history the record may match,
like the closed ones; blocking is still decided on the pull request's own alerts. Simulated on
#37's real alerts: the false "unused" goes, the real new alert stays. - SonarCloud failed main on the coverage of new code (78.6% against 80%) after the HTTPS-only
law fetch:fetch_htmlwithout a client of its own and the redirect limit had no test. Both
are now exercised against real servers on 127.0.0.1: a plain HTTP server that is never served
in clear, and an HTTPS server, with a certificate made byopensslfor the test, whose
redirects loop until the limit stops them. - Security Posture judged every pull request by main's alerts. It listed the alerts without
aref, so GitHub answered for main, and its wait compared the scanners' runs with the merge
commit, which none of them carries: on mailradar #39 it waited two seconds. It now waits for
the runs of the branch head, Bandit included, and reads the open alerts ofrefs/pull/N/merge;
the closed ones come from that ref and from main, so an exception main holds as settled is not
called unused on a ref with no history. Checked against a recent pull request of each Radar
with the real tool: the same verdict as main in all six. - The texts of the law arrived in clear. Cellar answers its https:// URL with a 303 to an
http:// one (recorded on 2026-10-11 for the GDPR), andfetch_htmlfollowed it, so the act
travelled without TLS and the SHA-256 printed beside a citation described bytes nothing had
protected. Redirects are now followed by hand and a hop to http:// is taken over https://,
which serves the same document byte for byte; if that fails, the fallback to EUR-Lex applies.
Checked live: the GDPR came back whole over two HTTPS requests. The test replays Cellar's
recorded 303. - The Saturday mutation run would have died on the NIS2 tests. mutmut copies only the
paths inalso_copyintomutants/, andtests/test_nis2_evidence.pyreadsdocs/, which
was not among them. Listed now, with a test that reads, from its own source, every root
path those tests use and checks that mutmut copies it. - The licence gate had never refused anything.
pip-licenses --fail-on="GPL;AGPL;LGPL"
compares whole licence strings, and no package declares just "GPL": they declare "GNU
Lesser General Public License v2 or later (LGPLv2+)" and the like. Measured on 2026-10-11
with chardet 5.2.0 installed next to cookieradar: exit 0 without--partial-match, exit 1
with it. The flag is there now; CookieRadar's own runtime dependencies pass it.
Security
- CVE-2026-88647 in gnutls28 recorded in
SECURITY-EXCEPTIONS.toml. A hostname
verification bypass with no fix in any Debian suite, found by the scan of the 2026.44
image; same chain as the five gnutls/krb5 entries of 2026-10-09 (libcups2t64, not
loaded by the headless shell), reviewed on 2026-11-30. - The gnutls28 and krb5 entries now say what was checked, not only what was read. The
six gnutls CVEs of 2026-10-08 were reproduced with the public PoC certificates against
Debian's own packages (trixie 3.8.9-3+deb13u4, sid 3.8.13-1) and had no upstream issue
anywhere on 2026-10-10; CVE-2026-107778 in krb5 gained Debian bug 1150461, a no-dsa for
trixie and a fix on the upstream 1.22 branch the same day. - **The...
CookieRadar 2026.44
Security
CVE-2026-95210,CVE-2026-95209,CVE-2026-95184,CVE-2026-67693(gnutls28) and
CVE-2026-107778(krb5) recorded inSECURITY-EXCEPTIONS.toml. Docker Scout opened
the five against 2026.43 on 2026-10-09, one of them critical;patchradar debianthe
same day says open in trixie, no fix in any suite, no Debian bug, and the tracker still
asks whether the four gnutls reports are valid. Neither library is in the base image:
playwright install-deps chromiumbrings them in for libcups2t64, which is the only
package depending on them. Measured in the published image, Playwright's fullchrome
links cups and so loads both; thechrome-headless-shellthat cookieradar runs — the
only mode that works in an image with no display — maps none of gnutls, krb5 or cups.
No TLS here goes through gnutls (httpx uses OpenSSL via_ssl, Chromium its BoringSSL)
and nothing uses Kerberos. Reviewed with the other Chromium dependencies by 2026-11-30.
Fixed
-
The report says when it was made and by what. The README calls it "evidence of
what the site did on a given day", and the file saved on 2026-10-09 for
www.repubblica.it carried no day: the only dates in it were the "Version of:" lines
under the legal citations, which exist only when something is cited, and nothing
named the CookieRadar that produced it. Every report now opens with
Audited on 2026-10-09 12:37 UTC with CookieRadar 2026.43, and the HTML file has a
<title>with the site's host — Rich's export template has none, so ten reports
open side by side were ten tabs calledreport.html. -
The law section no longer says "no cookie banner was found" under a report that
says "Banner found". Measured on 2026-10-09 on it.wikipedia.org, www.subito.it and
www.muenchen.de: every session printed the tick, neither button was clicked, and the
UNVERIFIED note denied the banner. Both lines came from one result. The heuristic
cannot tell a Didomi notice from a footer link called "Dichiarazione sui cookie"
(Wikipedia's match), and the note now says what it knows: something that looks like
a banner was found, and no refusal control was recognised on it. -
A page that never goes idle keeps its HTTP status.
page.goto(wait_until= "networkidle")returns the response only when the wait succeeds; when the page keeps
the network busy it raises a timeout and the status, which had arrived in the first
second, was lost with it. Measured on 2026-10-09: zalando.it answered 403 from its
edge with a full page whose scripts never went quiet, so the status was None,
page_not_servedsaw nothing wrong, and the audit said "0 trackers before consent,
UNVERIFIED" about an error page it never had; enel.it and poste.it, served with 200
and never idle, showed no status in any session. The status is now taken from the
main document's response as it arrives, last hop of a redirect, andgotoonly
confirms it.tests/site/forbidden_busy.html, served with 403 by the test server
under/forbidden/, reproduces the zalando.it case. -
TrustArc and Usercentrics refusals are clicked, and so are "Non accetto",
"Continua senza accettare" and "Solo gli essenziali". Measured on 2026-10-09:
enel.it and poste.it (TrustArc) offer refusal in one click —#truste-consent-required
"Continua senza accettare", whose banner says it leaves only technical cookies, and
#truste-consent-required2"Non accetto" — and the audit clicked "Accetta", found no
refusal, and reported "Accept was applied and no refusal control was found" with
GDPR art. 4(11) cited against both. zalando.it (Usercentrics) renders its banner in a
shadow root with no ids; the refusal is[data-testid="uc-deny-all-button"]"Solo gli
essenziali". The three banners are intests/site/, captured from the real pages
that day, and the sessions now click the refusal on each and the acceptance on none. -
An error page in any of the three sessions means NOT MEASURED.
page_not_served
read the pre-consent status only. In the batch run of 2026-10-09 www.zalando.it got no
response at all in the pre-consent session and the edge's 403 page in the post-reject
one — with a Usercentrics banner over it and Google Tag Manager loading once "Solo gli
essenziali" was clicked — and the verdict came out as VIOLATION, 2 trackers new after
rejection, assembled from one session that saw nothing and one that saw an error page.
The verdict rests on session 3; an error page there leaves it nothing to rest on. The
report now names the session that got the error page, and the other sessions' tables
say that no verdict rests on them.
Changed
- The test matrix runs on
3.15-devtoo, as an experimental row. 3.15.0 final is
due 2026-10-09 (PEP 790) and the dependencies the suite stands on already ship
cp315 wheels;actions/setup-pythonresolves3.15-devto the newest build, rc.3
today and the final once it is out. The row is anincludewith
continue-on-error, so it may fail without failing the run or the required
Testscheck; the classifiers still stop at 3.14 until 3.15 is final. patchradar
has had the same row since 2026-08-20.
CookieRadar 2026.43
Added
-
CVE-2026-77214recorded inSECURITY-EXCEPTIONS.toml. Docker Scout and Snyk
report the heap over-read in expat'sXML_ParseBuffer;patchradar debianon
2026-10-08 says fixed in sid only (2.9.0-1), open in trixie, bookworm and forky, no
Debian bug yet. Recorded the way the other expat entries are, to be reviewed by
2026-11-30, with who loads the systemlibexpat1in the image: fontconfig and Mesa,
for their own configuration files. CPython'spyexpatcarries its own copy, and
cookieradar parses no XML. -
The files the build is told to include are checked to be there. apkradar lost its
LICENSEout of the working tree on 2026-10-04 and the loss reachedmain: pyproject
names the file, sopython -m buildfailed withLicense file does not exist: LICENSE,
and the PyPI publish and the image went down with it. cookieradar lost its own a few
hours later, during a run of the suite. Neither suite noticed, because neither looked.What removes them is still not known, and these cases do not explain it. They stop it
reaching a commit, which is the part that can be fixed without knowing.The expectation is read out of the declarations rather than written down as
LICENSE,
because the five Radar do not declare it the same way: patchradar states its licence as
text and only its Dockerfile names the file, the other four name it in pyproject, and of
those apkradar and mailradar do not copy it into the image. So two cases — every file
pyproject names, and every path the Dockerfile copies — and between them each repository
is covered, three of them twice.Checked by moving the file aside in all five: it fails where it should and passes where
the declaration genuinely does not name it, and pointing pyproject at a file that is not
there fails too.
Fixed
-
Two defences in
release.shthat the tests did not actually measure. Found by
mutating the script rather than by reading it.Replacing the existing-tag check's
failwith anechoof the same words left every
case green: the script carried on, bumped, committed, and only then didgit tag
refuse the tag that already existed. The release was still refused — one commit too
late, which is the opposite of what the script promises, that a refusal leaves the
version file modified and nothing else. The cases now check that it did not commit on
its way to refusing.And
git push --atomicwas not measured at all; two separate pushes passed. With two
pushesmainarrives and the tag does not, so the repository carries a version bump
that no release and no published artifact corresponds to — and the tag that would
produce them cannot be pushed afterwards either, because the version it would be
given is by then "already the current version". Apre-receivehook on the test
remote now refuses tags, which is the way to make the second half fail on demand.Both mutations fail now, along with the three that already did.
-
The PyPI wait allows a margin once the index answers. apkradar's Docker build
failed on 2026-10-03 withNo matching distribution foundfifteen seconds after
the wait had reported the version available — 16:31:21 against 16:31:36. The poll is
not wrong and not enough: it establishes that the file is reachable from the runner,
while the build container, multi-platform through buildx, resolves the index again
and can reach an edge still serving the old one.This is not the
sleep 60that stood in that step before polling and lost the race
twice. That was a guess about how long publishing takes, made before knowing
anything; this waits for the fact first and then allows a bounded margin for it to
propagate, and says so in the log when it uses one.It narrows the window; it does not close it. What closes it is not asking the index
during the build at all — which is what exeradar's Dockerfile already does, with
pip install /app/src, and why exeradar has no wait script and did not hit this.
cookieradar and patchradar are one word from that (_SOURCE=local); apkradar and
mailradar would need the build argument added. That is the follow-up.Three cases hold the margin, and they were needed twice over. The two cases that
already drove this script pass the retry interval as zero so they stay fast, and the
new default made every success wait 45 seconds past their timeout — so the suite was
red in all four repositories until those two were told to ask for no grace. Telling
them that alone would have left the margin itself unmeasured, which is the shape of
defect this script was written to fix in the first place. So: one case times a
two-second grace and checks the log says why it waited, one checks that no grace
waits for nothing and claims nothing, and one measures the default without the
suite paying 45 seconds for it — started with no grace argument, the script must
still be running three seconds after the index answered. All four ways of undoing
the margin were checked against them: the default set to zero, the wait removed
while the log still claims it, the log removed while the wait still happens, and the
guard removed so zero waits anyway. -
batchno longer crashes on a site it could not see. The verdict was turned into
a label through a table with no entry for ERROR, so a site answering 403, or a
bot-management challenge answering 200, ended in❌ Error: <ExitCode.ERROR: 3>, and
the report asked for with-owas never written. It now reads⛔ NOT MEASUREDwith
the reason and no tracker counts — asauditalready did — and the report is saved. -
Auditing a site that is itself on the tracker list no longer finds it a tracker of
itself.linkedin.com,facebook.com,bing.com,tiktok.com: the site's own
document was counted as the trackerlinkedin.comin every session, so the verdict
was always VIOLATION "persists from pre-consent". Requests to the audited site are
first-party, which the README already put out of scope, andtrackersis again the
subset ofexternalthat its own comment says it is. -
The accept selector
button[id*='agree']no longer clicks "disagree". A substring
match: Didomi'sdidomi-notice-disagree-buttonsits next to its agree button, and
the selectors are tried before the labels, so the post-accept session rejected and
was recorded as accepted. -
The site is the one the browser landed on, spelled as the browser spells it. The
host compared against every request came from the address typed:müller.deagainst
requests toxn--mller-kva.de, orexample.comagainst a redirect toexample.it,
and every resource of the site came out as an unknown external host. The typed host
is now converted to IDNA, and the main frame's navigation — redirects included —
decides the site, in Chromium's own spelling (which also coversß, where Python's
IDNA 2003 codec and Chromium disagree). Checked against real Chromium on local pages. -
No provision is cited for a challenge page.
notes_ofknew that an error page is
not the site and did not know the same of a challenge served with 200, so the report
said NOT MEASURED and, under the provisions, "no cookie banner was found". Both now get
their own NOT MEASURED note, andfindings_ofdraws no finding from either: a reject
button on a page that is not the site no longer leads to citing a violation. -
normalize_urlrefuses an address with no host.https://passed, the browser
refused to navigate, and an invalid address exited 2 (UNVERIFIED) instead of 3. -
An unreadable
--trackersdatabase exits 3, not 1. OnlyValueErrorwas caught,
so a.enothat could not be read escaped as a traceback with exit status 1 — which
a pipeline reads as VIOLATION. -
batch -oreport names no longer collide where the filesystem ignores case.
Example.com.txtandexample.com.txtare one file on Windows and macOS, and the
second report overwrote the first. -
A trackerdb block left open is kept when another block follows it, not only at the
end of the file, as the parser's docstring already promised. -
Shell scripts are checked out with LF everywhere (
.gitattributes). With
core.autocrlf=truea Windows checkout wroterelease.shandwait_for_pypi.shwith
CRLF, and bash stopped atset: pipefail\r: invalid option name. -
pytest, as the README says to run it, no longer ends in three errors. The
benchmarks need the CodSpeed runner, which onlycodspeed.ymlinstalls; without it
they are now skipped with that reason. Andpyproject.tomlno longer sends the reader
totests/test_benchmark_contract.py, a file that never existed. -
README:
audit --trackersis in the command reference, the version examples use
the current scheme, andbatch's NOT MEASURED is described.
Changed
-
The race with PyPI is closed rather than narrowed: the released image no longer asks
the index.docker.ymlinstalledcookieradar==<the new version>from PyPI and
polled the index first to make that work. The poll is not wrong and not enough: it runs
on the runner, while the multi-platform build resolves the index again, per platform,
from whichever edge answers — apkradar lost that race on 2026-10-03 fifteen seconds
after its poll had succeeded. Nothing that waits can close it. Not asking does, so the
image is built from the source the tag points at.localis also the branchtests.ymlalready builds on every run, with the smoke test
behind it, so the release moves onto the better exercised of the two paths rather than
onto an untried one. This repository is the only one of the five where that was already
true; in the other four the only thing that built the image automatically was the
workflow that publishes it.Two things follow. A dispatched rebuild now chec...
v2026.42
What's Changed
- fix(ci): a fixed tag for the image Snyk scans by @maksimtech in #20
- fix(ci): mutation testing runs again — also_copy for what the tests read by @maksimtech in #22
- ci: pin the runners to ubuntu-26.04, benchmarks to 24.04 by @maksimtech in #21
Full Changelog: v2026.41...v2026.42
v2026.41
Full Changelog: v2026.40.1...v2026.41
v2026.40.1
What's Changed
- build(deps): bump actions/upload-artifact from 4 to 7 by @dependabot[bot] in #19
Full Changelog: v2026.40...v2026.40.1