critical

CVE

CVE-2026-123272, CVE-2026-123271, CVE-2026-113663, CVE-2026-123783, CVE-2026-61536, CVE-2026-110930, CVE-2026-73098, CVE-2026-121489, CVE-2026-73099, CVE-2026-73100, CVE-2026-73101, CVE-2026-73116, CVE-2026-73117, CVE-2026-96677, CVE-2026-73095, CVE-2026-121819, CVE-2026-121820, CVE-2026-121821, CVE-2026-121822, CVE-2026-121823, CVE-2026-121824, CVE-2026-123561, CVE-2026-123562, CVE-2026-123563, CVE-2026-123564, CVE-2026-123565, CVE-2026-123566, CVE-2026-123567, CVE-2026-123568, CVE-2026-123569, CVE-2026-124492, CVE-2026-124493, CVE-2026-124494, CVE-2026-121623, CVE-2026-118515, CVE-2026-123792, CVE-2026-123824, CVE-2026-123488, CVE-2026-123489, CVE-2026-123490, CVE-2026-123491, CVE-2026-124427, CVE-2026-124429, CVE-2026-124430, CVE-2026-124437, CVE-2026-124439, CVE-2026-123551, CVE-2026-123549, CVE-2026-123550, CVE-2026-124450, CVE-2026-124469, CVE-2026-124445, CVE-2026-124467

CWE

CWE-346, CWE-78, CWE-94, CWE-470, CWE-22, CWE-611, CWE-200, CWE-400, CWE-1321, CWE-89, CWE-287, CWE-863

Affected Surface

  • Go MCP clients built on `github.com/mark3labs/mcp-go` `< v1.2.1` (`CVE-2026-123272` and `GO-2026-6911`, CVSS 8.0 High) and `github.com/diillson/chatcli` `< v1.215.2` (`CVE-2026-123271`, CVSS 8.0 High)
  • AI agent workflows using Python `ag2` `>= 0.13.2, < 1.1.1` or `autogen` `>= 0.13.2, <= 0.14.1` (`CVE-2026-113663`, CVSS 8.8 High), `skyvern` `<= 1.0.56` (`CVE-2026-123783`, CVSS 8.8 High), `banks` `<= 2.4.2` (`CVE-2026-61536` / `GHSA-64vx-6h2c-rjh7`, CVSS 7.5 High), and OpenAI Codex Desktop for Windows (`CVE-2026-110930`, CVSS 8.0 High)
  • Java microservices and identity servers using `org.apache.juneau:*` `<= 9.2.0` (`CVE-2026-73098`, `CVE-2026-121489`, `CVE-2026-73099`, `CVE-2026-73116`, `CVE-2026-73117`, `CVE-2026-96677`, `CVE-2026-73095`), `com.linecorp.armeria:*` `< 1.43.0` (`CVE-2026-121819` through `CVE-2026-121824`), and OpenIdentityPlatform `OpenAM` `< 16.1.4` (`CVE-2026-123551` CVSS 10.0 Critical, `CVE-2026-123549` CVSS 9.8 Critical, `CVE-2026-123550` CVSS 9.8 Critical)
  • JavaScript/SvelteKit applications using npm `devalue` `< 5.9.3` (`CVE-2026-123561` through `CVE-2026-123569`) or `suneditor` `< 3.4.0` (`CVE-2026-124492` through `CVE-2026-124494`), PHP applications using `symfony/serializer`, `symfony/yaml`, or `symfony/config` prior to `5.4.56`, `6.4.50`, `7.4.23`, and `8.1.11` (`CVE-2026-121623`, `CVE-2026-118515`, `CVE-2026-123792`, `CVE-2026-123824`), `craftcms/cms` `< 4.18.9` / `< 5.11.2` (`CVE-2026-123488` to `123491`, `CVE-2026-124427` to `124430`), or `openemr/openemr` `< 8.3.0` / `< 8.4.0` (`CVE-2026-124450`, `CVE-2026-124469`, `CVE-2026-124445`, `CVE-2026-124467`), and .NET applications using NuGet `Scriban` `< 7.5.3` (`CVE-2026-124437`, `CVE-2026-124439`)

TL;DR

Between 5 and 8 November 2026, NVD and the GitHub Advisory Database published security advisories across three application security fronts:

  1. MCP OAuth token theft and AI agent sandbox escapes (mark3labs/mcp-go, chatcli, AG2, Skyvern, banks, and Codex Desktop):
    • CVE-2026-123272 (mark3labs/mcp-go < v1.2.1, CVSS 8.0 High, plus GO-2026-6911) and CVE-2026-123271 (chatcli < v1.215.2, CVSS 8.0 High): When discovering OAuth 2.0 Protected Resource Metadata (RFC 9728) from /.well-known/oauth-protected-resource, OAuthHandler.fetchServerMetadata() in mcp-go enforced the RFC 9728 section 3.3 resource binding check only when the metadata URL came from a WWW-Authenticate header, skipping it on the well-known path. A malicious MCP server could declare a victim service’s URL in its "resource" field, causing the client to request an RFC 8707 audience-bound token for the victim service and hand it to the attacker’s MCP server.
    • CVE-2026-113663 (ag2 < 1.1.1 / autogen <= 0.14.1, CVSS 8.8 High): SandboxShellTool in restricted mode (allowed=[...] or readonly=True) validated only the head command and missed &, \n, \r, and < before passing the raw string to sh -c, while allowing default readonly commands (find . -exec ..., env touch x, sort -o x, git diff --output=x) to write files and execute subcommands.
    • CVE-2026-123783 (skyvern <= 1.0.56, CVSS 8.8 High) and CVE-2026-61536 (banks <= 2.4.2, CVSS 7.5 High): Skyvern’s code_block_safety.py AST blocklist omitted types.CodeType, allowing workflow code blocks to construct raw Python bytecode objects and bypass BLOCKED_ATTRS. In banks, CompletionExtension parsed Tool JSON objects from rendered {% completion %} prompt text and passed tool.import_path straight to importlib.import_module() + getattr() without an allowlist, bypassing the Jinja SandboxedEnvironment fix from 2.4.2.
  2. Apache Juneau 9.2.1 (8 CVEs including seven CVSS 10.0 Critical) and LINE Armeria 1.43.0 (6 CVEs): Juneau <= 9.2.0 called Class.forName(name, true, ...) when deserializing java.lang.Class properties (running static initializers of arbitrary classes on the classpath, CVE-2026-73098), allowed path traversal outside DirectoryResource roots (CVE-2026-121489), replaced GET bodies via a default-enabled ?content= query parameter (CVE-2026-73099), and forwarded Authorization and Cookie headers across cross-origin redirects in RestClient (CVE-2026-73116). LINE Armeria < 1.43.0 patched inconsistent path normalization bypassing route access controls (CVE-2026-121823, CVSS 8.7 High), unbounded SAML HTTP-Redirect DEFLATE decompression (CVE-2026-121820, CVSS 8.7 High), and Thrift message allocation DoS (CVE-2026-121824, CVSS 8.7 High).
  3. Svelte devalue 5.9.3 (CVE-2026-123561 to 123569), SunEditor 3.4.0 (CVE-2026-124492 to 124494), Symfony 5.4.56 / 6.4.50 / 7.4.23 / 8.1.11 (CVE-2026-121623, CVE-2026-118515), Craft CMS 5.11.2 / 4.18.9 (CVE-2026-123488, CVE-2026-124427, CVE-2026-124430), and .NET Scriban 7.5.3 (CVE-2026-124437, CVE-2026-124439): In Symfony’s Serializer, DataUriNormalizer::denormalize() matched ^data: case-insensitively (preg_match('/^data:.../i')) and then passed the raw string to new \SplFileObject($data), opening "DATA:,/../../etc/passwd" as a local file (CVE-2026-121623, CVSS 8.2 High), while UploadedFile properties fell through to ObjectNormalizer (CVE-2026-118515, CVSS 8.4 High). Craft CMS 5.11.2 / 4.18.9 patched an unauthenticated stacked-query SQL injection in the element search endpoint (CVE-2026-123488, CVSS 9.8 Critical), an anonymous GraphQL scope bypass via prev/next hops (CVE-2026-124427, CVSS 9.8 Critical), and a Twig signing-oracle RCE via app/render-elements + login returnUrl (CVE-2026-124430, CVSS 8.8 High).

Affected packages and frameworks

EcosystemPackageCVE / GHSASeverityAffected versionsFixed versions
Gogithub.com/mark3labs/mcp-goCVE-2026-123272, GO-2026-6911High (8.0)< v1.2.1v1.2.1
Gogithub.com/diillson/chatcliCVE-2026-123271High (8.0)< v1.215.2v1.215.2
PyPIag2, autogenCVE-2026-113663 (GHSA-2h4w-29cv-2cmv)High (8.8)ag2 >= 0.13.2, < 1.1.1; autogen >= 0.13.2, <= 0.14.1ag2 1.1.1
PyPIskyvernCVE-2026-123783High (8.8)<= 1.0.56Upstream patch after 1.0.56
PyPIbanksCVE-2026-61536 (GHSA-64vx-6h2c-rjh7)High (7.5)<= 2.4.22.4.3
Mavenorg.apache.juneau:juneau-*CVE-2026-73098, 121489, 73099, 73100, 73101, 73116, 73117, 96677, 73095Critical (10.0) / High (8.7)<= 9.2.09.2.1
Mavencom.linecorp.armeria:armeria*CVE-2026-121819 through CVE-2026-121824 (GHSA-xqhm-4ph4-423v, GHSA-6hgp-j5h7-33m3, et al.)High (8.7) / Medium< 1.43.01.43.0
npmdevalue, suneditorCVE-2026-123561..123569 (devalue); CVE-2026-124492..124494 (suneditor)Critical (9.3) / High (7.5)devalue < 5.9.3; suneditor < 3.4.0devalue 5.9.3; suneditor 3.4.0 (3.4.3 latest)
Packagistsymfony/symfony, symfony/serializer, symfony/yaml, symfony/configCVE-2026-118515, CVE-2026-121623, CVE-2026-123792, CVE-2026-123824High (8.7, 8.4, 8.2) / Medium< 5.4.56, 6.x < 6.4.50, 7.x < 7.4.23, 8.x < 8.1.115.4.56, 6.4.50, 7.4.23, 8.1.11
Packagistcraftcms/cmsCVE-2026-123488..123491, CVE-2026-124427, 124429, 124430Critical (9.8) / High (8.8)4.0.0..4.18.8; 5.0.0..5.11.14.18.9, 5.11.2 (5.12.0.1 latest)
Packagistopenemr/openemrCVE-2026-124450 (GHSA-wf4w-mx2w-j4vv), CVE-2026-124469 (GHSA-cg8c-2v5q-29g7), CVE-2026-124445 (GHSA-xg3m-9pfq-c8r6), CVE-2026-124467 (GHSA-p4v7-crwp-qq7q)Critical (9.9, 9.8) / High (8.8)< 8.3.0 (124450, 124469); < 8.4.0 (124445, 124467)8.3.0, 8.4.0 (8.5.0 latest)
Mavenorg.openidentityplatform.openam:openam-*CVE-2026-123551 (GHSA-5c3q-hp86-xjr8), CVE-2026-123549 (GHSA-jv8f-m88g-h6mm), CVE-2026-123550 (GHSA-3j2f-qhrc-rjfj)Critical (10.0, 9.8)14.0.0..16.1.316.1.4
NuGetScriban, Scriban.SignedCVE-2026-124437 (GHSA-95w2-9mjq-6c74), CVE-2026-124439 (GHSA-mvmw-9jxc-7pr2)High (7.5)< 7.5.37.5.3

Technical analysis

1. CVE-2026-123272 (mark3labs/mcp-go) and CVE-2026-123271 (chatcli): RFC 9728 Protected Resource Metadata token-audience hijacking

In the MCP OAuth 2.0 authorization specification, an MCP client discovers a server’s authorization requirements by fetching RFC 9728 Protected Resource Metadata (PRM), either from a URL advertised in WWW-Authenticate: Bearer resource_metadata="..." or from the well-known URI /.well-known/oauth-protected-resource<path>. RFC 9728 section 3.3 requires the client to verify that the "resource" identifier inside the returned JSON document matches the MCP server URL the client actually connected to before passing that "resource" value as the RFC 8707 resource audience indicator in token requests.

In mark3labs/mcp-go < v1.2.1 (commit 90d247ed8d45), OAuthHandler.fetchServerMetadata() in client/transport/oauth.go only ran the binding check when the metadata URL came from WWW-Authenticate (fromChallenge == true), trusting the well-known response unconditionally:

--- a/client/transport/oauth.go
+++ b/client/transport/oauth.go
@@ -799,14 +799,12 @@ func (h *OAuthHandler) fetchServerMetadata(ctx context.Context) (metadataDiscove
-	if fromChallenge {
-		if metadata.Resource == "" {
-			return metadataDiscoveryResult{}, fmt.Errorf("protected resource metadata from %s is missing required resource field", mdURL)
-		}
-		if !matchesResourceURI(metadata.Resource, h.baseURL) {
-			return metadataDiscoveryResult{}, fmt.Errorf("protected resource metadata resource %q does not match server %q", metadata.Resource, h.baseURL)
-		}
+	if metadata.Resource == "" {
+		return metadataDiscoveryResult{}, fmt.Errorf("protected resource metadata from %s is missing required resource field", mdURL)
+	}
+	if !matchesResourceURI(metadata.Resource, h.baseURL) {
+		return metadataDiscoveryResult{}, fmt.Errorf("protected resource metadata resource %q does not match server %q", metadata.Resource, h.baseURL)
 	}

Without that check, an untrusted MCP server at https://evil-mcp.example.com could serve /.well-known/oauth-protected-resource declaring "resource": "https://internal-api.corp.example.com" and pointing authorization_servers at the organization’s shared SSO provider. When the user completed the OAuth flow, the authorization server minted an access token scoped to https://internal-api.corp.example.com, and mcp-go delivered that token directly to evil-mcp.example.com (CVE-2026-123271 fixed the identical flaw plus cross-origin SSE endpoint event hijacking in chatcli v1.215.2, and GO-2026-6911 fixed mcp-go forwarding bearer tokens across unvalidated HTTP redirect Location headers).

2. AI agent sandbox escapes in AG2 1.1.1 (CVE-2026-113663) and banks 2.4.3 (CVE-2026-61536)

In ag2 < 1.1.1 (commit 3cd86726bb73, GHSA-2h4w-29cv-2cmv), ShellAdapter in ag2/tools/sandbox/adapter/shell.py offered a restricted mode (allowed=[...] or readonly=True) to constrain what commands an LLM agent could run. However, contains_shell_operator() in ag2/tools/sandbox/filter.py checked only for ;, |, &&, ||, >, >>, `, and $(, omitting background execution (&), newline (\n), carriage return (\r), and process substitution (<(...)), and then handed the command string to sh -c. Furthermore, the default READONLY_COMMANDS tuple included find, env, sort, file, and git, allowing an agent in readonly=True mode to run find . -exec sh -c ... \;, env python3 -c ..., or git status & curl .... Commit 3cd86726bb73 parses restricted commands into argv via shlex, removes find, env, file, sort, and git from READONLY_COMMANDS, and executes the validated argv directly without invoking a shell.

In banks <= 2.4.2 (CVE-2026-61536 / GHSA-64vx-6h2c-rjh7), CompletionExtension._execute_tool_calls() in src/banks/extensions/completion.py extracted Tool JSON metadata embedded by the {% tool %} tag inside a rendered {% completion %} block and resolved tool.import_path dynamically:

# banks <= 2.4.2: src/banks/extensions/completion.py
module_name, func_name = tool.import_path.rsplit(".", 1)
module = importlib.import_module(module_name)
func = getattr(module, func_name)

Because Tool definitions are parsed as plain JSON from the rendered prompt text after Jinja template evaluation finishes, switching banks to SandboxedEnvironment in 2.4.2 (CVE-2026-44209) did not stop prompt-injected JSON from specifying "import_path": "os.system" or "subprocess.getoutput". When the model returned a matching tool_calls entry, banks imported os and executed os.system(...) with the LLM’s arguments.

3. Symfony DataUriNormalizer case-sensitivity mismatch (CVE-2026-121623), UploadedFile injection (CVE-2026-118515), and Craft CMS 5.11.2 (CVE-2026-123488, CVE-2026-124430)

Symfony’s DataUriNormalizer converts data: URI strings into SplFileInfo, SplFileObject, or Symfony\Component\HttpFoundation\File\File instances during deserialization. Two bugs in src/Symfony/Component/Serializer/Normalizer/DataUriNormalizer.php turned that normalizer into an arbitrary local file reader:

  1. Case-insensitive regex vs. case-sensitive PHP stream wrapper (CVE-2026-121623, commit a0c551861ac2): denormalize() validated the input with preg_match('/^data:...$/i', $data) (using the /i case-insensitive flag) and then passed $data unchanged to new \SplFileObject($data). In PHP’s stream layer, however, the data:// wrapper is registered in lowercase only; an uppercase scheme such as DATA:,/../../etc/passwd is not recognized as a stream wrapper and is opened as a relative filesystem path (where /.. climbs out of DATA:,/ to / and reads /etc/passwd):
    --- a/src/Symfony/Component/Serializer/Normalizer/DataUriNormalizer.php
    +++ b/src/Symfony/Component/Serializer/Normalizer/DataUriNormalizer.php
    @@ -99,6 +99,9 @@ public function denormalize($data, string $type, ?string $format = null, array $
                 throw NotNormalizableValueException::createForUnexpectedDataType('The provided "data:" URI is not valid.', $data, ['string'], $context['deserialization_path'] ?? null, true);
             }
    +
    +        // PHP opens a "data:" URI as a local path when its scheme is not lowercase
    +        $data = 'data:'.substr($data, 5);
  2. SplFileInfo subclass fallthrough to ObjectNormalizer (CVE-2026-118515, commit 72ece5bb3084): DataUriNormalizer::SUPPORTED_TYPES listed only the exact class names SplFileInfo, SplFileObject, and File. If a DTO property was typed as a subclass such as Symfony\Component\HttpFoundation\File\UploadedFile, DataUriNormalizer declined it and ObjectNormalizer instantiated new UploadedFile($path, $originalName, ..., test: true) directly from attacker-supplied JSON keys, pointing the UploadedFile object at any file on the server disk.
  3. Craft CMS 4.18.9 / 5.11.2 (CVE-2026-123488, CVE-2026-124427, CVE-2026-124430): In craftcms/cms, the element search endpoint concatenated the attribute parameter directly into SQL where emulated PDO prepares allowed stacked statements to overwrite users password hashes or admin flags (CVE-2026-123488, CVSS 9.8 Critical); anonymous GraphQL clients could chain prev and next field hops from a single visible entry to read entries in withheld sections (CVE-2026-124427, CVSS 9.8 Critical); and non-admin control panel users could use app/render-elements as an HMAC signing oracle to mint a signed Twig object template and submit it as the login returnUrl, triggering unsandboxed View::renderObjectTemplate() RCE (CVE-2026-124430, CVSS 8.8 High).

4. Apache Juneau 9.2.1 (CVE-2026-73098), Svelte devalue 5.9.3 (CVE-2026-123561), SunEditor 3.4.0 (CVE-2026-124494), and .NET Scriban 7.5.3 (CVE-2026-124439)

  • Apache Juneau 9.2.1 (commit 81677c11fb and b91bce1617): When deserializing a string into a bean property of type java.lang.Class, Juneau’s default ClassSwap called Class.forName(name) with initialize = true, executing the static {} initializer block of any class named by the client (CVE-2026-73098, CVSS 10.0 Critical). In juneau-microservice-core, DirectoryResource failed to canonicalize request paths against its configured root (CVE-2026-121489, CVSS 10.0 Critical), RestClient replayed Authorization, Cookie, and X-API-Key headers across cross-origin HTTP redirects (CVE-2026-73116, CVSS 10.0 Critical), and the REST server accepted ?content=<body> on GET requests by default (CVE-2026-73099, CVSS 10.0 Critical).
  • Svelte devalue 5.9.3 (compare v5.9.2...v5.9.3) and SunEditor 3.4.0 (CVE-2026-124493, CVE-2026-124494): Used by SvelteKit for server-to-client and form-action serialization, devalue 5.9.3 fixes nine CVEs: non-string keys in null-prototype objects coercing to "__proto__" after the key check in parse and unflatten (CVE-2026-123561), custom ArrayBuffer revivers bypassing typed-array allocation bounds (CVE-2026-123563), unhandled promise rejections crashing Node.js in stringifyAsync (CVE-2026-123562), and sparse-array CPU/memory amplification in uneval (CVE-2026-123564 to 123569). In SunEditor < 3.4.0, html.clean()’s _normalizeURL() stripped named HTML entities such as &colon; to an empty string before _isSafeURL() checked the scheme, letting href="javascript&colon;alert(1)" pass sanitization and execute when clicked (CVE-2026-124494, CVSS 9.3 Critical).
  • .NET Scriban 7.5.3 (commits 466aeb2f0cf6 and 6b9c69bf9178), OpenAM 16.1.4 (CVE-2026-123551), and OpenEMR 8.3.0 / 8.4.0 (CVE-2026-124450, CVE-2026-124469): In Scriban < 7.5.3, postfix member/indexer/call chains bypassed ExpressionDepthLimit and object.eval bypassed RecursiveLimit, allowing untrusted templates to trigger an uncatchable .NET StackOverflowException that terminates the host worker process (CVE-2026-124439), while builtins such as math.format, regex.replace, and string.escape skipped the LimitToString allocation cap (CVE-2026-124437). In OpenIdentityPlatform OpenAM 14.0.0 before 16.1.4, unauthenticated callers could bypass the agent-only check on the notification WebSocket to read live session tokens (CVE-2026-123551, CVSS 10.0 Critical), assign restricted group memberships during legacy self-registration (CVE-2026-123549, CVSS 9.8 Critical), and submit unsigned SAML v2.0 assertions on the ECP (PAOS) and HTTP-Artifact bindings when WantAssertionsSigned was false (CVE-2026-123550, CVSS 9.8 Critical). In OpenEMR < 8.3.0 / < 8.4.0, the Fax/SMS disposeDocument action allowed authenticated low-privilege users to read sqlconf.php or write arbitrary files via file_path (CVE-2026-124450, CVSS 9.9 Critical), DrugSalesService::sellDrug interpolated prod[N][warehouse] into an ORDER BY clause (CVE-2026-124469, CVSS 9.9 Critical), OneTimeAuth.php ignored enforce_onetime_use and max_access_count (CVE-2026-124445, CVSS 9.8 Critical), and library/ESign/Router.php passed an unsanitized module parameter into require_once '<module>/Controller.php' (CVE-2026-124467, CVSS 8.8 High).

IOCs

Check application logs, WAF telemetry, and dependency lockfiles for these exploitation and version indicators:

  • Symfony DataUriNormalizer requests (CVE-2026-121623): JSON or XML payloads supplying uppercase or mixed-case DATA:,/... strings (^DATA: or ^Data:) containing ../ traversal sequences to endpoints that deserialize SplFileInfo, SplFileObject, or File properties.
  • Craft CMS requests (CVE-2026-123488, CVE-2026-124430): HTTP requests to the element search action with SQL metacharacters in attribute, or non-admin requests to actions/app/render-elements paired with signed returnUrl parameters.
  • SunEditor stored HTML (CVE-2026-124494): Stored anchor or media attributes containing javascript&colon;, javascript&newline;, or javascript&Tab;.

Detection and remediation

# 1. Check Go, Python, Maven/Gradle, npm, Composer, and NuGet lockfiles for vulnerable versions
rg -n 'mark3labs/mcp-go|diillson/chatcli|ag2\b|autogen\b|skyvern|banks\b|org\.apache\.juneau|com\.linecorp\.armeria|"devalue"|"suneditor"|symfony/(symfony|serializer|yaml|config)|craftcms/cms|Scriban' \
  go.mod pyproject.toml requirements*.txt uv.lock pom.xml build.gradle* package-lock.json pnpm-lock.yaml yarn.lock composer.lock *.csproj packages.lock.json 2>/dev/null

# 2. Search Symfony codebases for DTOs deserializing SplFileInfo, SplFileObject, File, or UploadedFile
rg -n 'SplFileInfo|SplFileObject|UploadedFile|DataUriNormalizer|XmlUtils::parse' src/ 2>/dev/null
  1. Upgrade MCP and AI agent packages: Update github.com/mark3labs/mcp-go to >= v1.2.1, chatcli to >= v1.215.2, ag2 to >= 1.1.1, and banks to >= 2.4.3. For skyvern <= 1.0.56, add "CodeType" to BLOCKED_ATTRS in skyvern/forge/sdk/workflow/code_block_safety.py until upgrading past 1.0.56.
  2. Upgrade Java, JavaScript, PHP, and .NET libraries: Update Apache Juneau (org.apache.juneau:*) to 9.2.1, LINE Armeria (com.linecorp.armeria:*) to 1.43.0, npm devalue to >= 5.9.3 and suneditor to >= 3.4.0 (3.4.3), Symfony (symfony/serializer, symfony/yaml, symfony/config, symfony/http-foundation, symfony/security-http) to 5.4.56, 6.4.50, 7.4.23, or 8.1.11, Craft CMS (craftcms/cms) to >= 4.18.9 or >= 5.11.2, and NuGet Scriban to >= 7.5.3.

Timeline

  • 2026-11-04 (oss-security 2026/11/04/7): Apache Juneau announces 9.2.1 fixing eight security vulnerabilities (CVE-2026-73095 through CVE-2026-121489).
  • 2026-11-05 (17:51 to 23:16 UTC): GitHub Advisory Database and NVD publish CVE-2026-61536 (banks 2.4.3), CVE-2026-123272 (mark3labs/mcp-go 1.2.1), CVE-2026-123271 (chatcli 1.215.2), CVE-2026-110930 (Codex Desktop), and CVE-2026-113663 (AG2 1.1.1).
  • 2026-11-06 (08:17 to 18:18 UTC): NVD publishes the six LINE Armeria 1.43.0 CVEs (CVE-2026-121819 to 121824), CVE-2026-123488 to 123491 (Craft CMS), CVE-2026-123783 (Skyvern), and the Symfony 5.4.56 / 6.4.50 / 7.4.23 / 8.1.11 CVEs (CVE-2026-121623, CVE-2026-118515, CVE-2026-123792, CVE-2026-123824).
  • 2026-11-08 (02:16 to 14:18 UTC): NVD publishes nine CVEs (CVE-2026-123561 through CVE-2026-123569) for Svelte devalue < 5.9.3, CVE-2026-124427 to 124430 for Craft CMS 5.11.2, CVE-2026-124437 and 124439 for .NET Scriban 7.5.3, and CVE-2026-124492 to 124494 for SunEditor 3.4.0.

From research to remediation

Check whether this pattern exists in your codebase

Turn this research into a remediation workflow. Scan dependencies and package manifests for similar supply-chain risk, then prioritize fixes with reachability context.

References