Security Daily Brief
20 events44 source observations
Lead stories
nltk: 18 security advisories
GitHub published 18 security advisories affecting nltk. Reported issues include: NLTK: ZIP package extraction follows in-root hardlinks and overwrites files outside the NLTK data root; NLTK: AllowlistUnpickler does not deny numpy submodule-local file-I/O sinks; NLTK corpus-reader view methods bypass the scoped root sandbox via intermediate-directory symlinks.
Read the analysisThe day in context
GitHub’s 18 NLTK advisories include data-root boundary failures; one requires a local actor able to write in a shared data root. A Miren Runtime advisory says unverified client certificates can bypass cluster authorization when the RPC listener is reachable. Rapid7 reports that Broadcom rated the vCenter authentication bypass 9.8 and provided updates. Todoist’s optional local HTTP service faces browser DNS rebinding. Mandiant and GTIG documented ARMORY CASTLE’s use of a patched BeyondTrust zero-day and subsequent vCenter activity. Cloudflare saw more vulnerability attempts than DDoS attacks against government election websites in its protection programs. AWS announced year-end security sessions. Operators should check component and network exposure separately.
An Axon advisory describes missing integrity checks on a downloaded executable; the stated attack requires compromise of the release pipeline or a maintainer token. Rapid7 attributes a custom SoftEther tool to Webworm with moderate confidence. Docker says it has begun publishing fixed Java4Shell and NGINX images, but fixed Temurin builds are not yet released; Java exposure also depends on hardware. Media report NGINX fixes. An OpenID4VP advisory warns that chain validation may accept forged credentials. AWS says most incident-response triage does not appear as cases, so case counts alone do not show all activity.
TechRadar reports credential-stealing malware in Microsoft’s official Python agent framework; the attack was stopped quickly, but victims may exist. An advisory warns that some MCP Python SDK clients could send credentials to an untrusted server’s chosen endpoint. Help Net Security says PyPI quarantined two malicious framework releases. GTIG observed a Russian espionage group impersonating employees of breached organizations to phish new targets. Other advisories describe a gosaml2 crash, Guard Core denial of service and missing authentication in k8s-mcp-server’s default transport. Teams should check versions, configurations and reachability.
Lead stories
nltk: 18 security advisories
Published:
GitHub published 18 security advisories affecting NLTK, covering distinct issues in data-directory boundaries, deserialization and local program execution. One advisory says package extraction can follow a hardlink planted inside the NLTK data root and overwrite a file outside it. Exploitation requires local write access to the data root and is also constrained by hardlink permissions or system protections. The advisory identifies published releases through 3.10.3 as affected and says an earlier downloader fix did not cover this extraction path. A separate advisory describes local code execution when prover9 or mace4 is configured with a relative binary location and an attacker can place a program in the resolved directory; default bare-name searches have a guard. Teams maintaining shared NLTK data directories or invoking these tools should review directory permissions and configuration separately. These findings should not be treated as one remote attack, and an earlier fix should not be assumed to cover every affected path.
nltk: 18 security advisories
Published:
GitHub published 18 security advisories affecting NLTK, covering distinct issues in data-directory boundaries, deserialization and local program execution. One advisory says package extraction can follow a hardlink planted inside the NLTK data root and overwrite a file outside it. Exploitation requires local write access to the data root and is also constrained by hardlink permissions or system protections. The advisory identifies published releases through 3.10.3 as affected and says an earlier downloader fix did not cover this extraction path. A separate advisory describes local code execution when prover9 or mace4 is configured with a relative binary location and an attacker can place a program in the resolved directory; default bare-name searches have a guard. Teams maintaining shared NLTK data directories or invoking these tools should review directory permissions and configuration separately. These findings should not be treated as one remote attack, and an earlier fix should not be assumed to cover every affected path.
Sources · 8
Miren Runtime: Authentication bypass in RPC layer: unverified TLS client certificates grant cluster superuser access
Published:
A GitHub advisory reports that the Miren Runtime RPC server requested TLS client certificates without verifying that they chained to the cluster certificate authority. It also bypassed role-based authorization checks for certificate identities. Consequently, anyone able to reach the RPC listener could present a forged certificate and obtain cluster superuser access, with the ability to manipulate applications, disks and other cluster resources. Exposure depends on network reachability: an internet-facing port is remotely accessible, while restricting it to a trusted network still leaves parties on that network able to reach it. The advisory says the fix verifies a presented certificate's chain before trusting it while still permitting callers without certificates to authenticate using tokens. Because unauthorized RPC access cannot be detected, Miren recommends redeploying all applications and rotating credentials referenced in environment variables. Operators should check the listener's exposure and their fix status, and assess credential rotation against the vendor's recommendation.
Miren Runtime: Authentication bypass in RPC layer: unverified TLS client certificates grant cluster superuser access
Published:
A GitHub advisory reports that the Miren Runtime RPC server requested TLS client certificates without verifying that they chained to the cluster certificate authority. It also bypassed role-based authorization checks for certificate identities. Consequently, anyone able to reach the RPC listener could present a forged certificate and obtain cluster superuser access, with the ability to manipulate applications, disks and other cluster resources. Exposure depends on network reachability: an internet-facing port is remotely accessible, while restricting it to a trusted network still leaves parties on that network able to reach it. The advisory says the fix verifies a presented certificate's chain before trusting it while still permitting callers without certificates to authenticate using tokens. Because unauthorized RPC access cannot be detected, Miren recommends redeploying all applications and rotating credentials referenced in environment variables. Operators should check the listener's exposure and their fix status, and assess credential rotation against the vendor's recommendation.
Todoist MCP Server: Browser DNS rebinding lets websites write and delete Todoist data through local MCP server
Published:
A GitHub advisory says the optional local HTTP service for Todoist MCP Server can allow a malicious website to invoke Todoist tools through browser-origin DNS rebinding, including tools that read, write or delete data. The finding concerns todoist-mcp-http in package version 10.3.1. It does not validate the request's Host or Origin or authenticate the browser caller; it validates the Todoist token configured on the server instead. Tools then run using the operator's API key, so binding to loopback by default does not eliminate the browser-origin risk. The advisory distinguishes the stdio entrypoint, which is not browser-reachable. Operators should establish whether they have enabled the optional HTTP mode and assess the local service and the permissions of its Todoist account. A deployment using only stdio is not the browser-reachable scenario described in the advisory. The finding depends on the HTTP service running and a user visiting a malicious site; it does not mean every installation has been accessed.
Todoist MCP Server: Browser DNS rebinding lets websites write and delete Todoist data through local MCP server
Published:
A GitHub advisory says the optional local HTTP service for Todoist MCP Server can allow a malicious website to invoke Todoist tools through browser-origin DNS rebinding, including tools that read, write or delete data. The finding concerns todoist-mcp-http in package version 10.3.1. It does not validate the request's Host or Origin or authenticate the browser caller; it validates the Todoist token configured on the server instead. Tools then run using the operator's API key, so binding to loopback by default does not eliminate the browser-origin risk. The advisory distinguishes the stdio entrypoint, which is not browser-reachable. Operators should establish whether they have enabled the optional HTTP mode and assess the local service and the permissions of its Todoist account. A deployment using only stdio is not the browser-reachable scenario described in the advisory. The finding depends on the HTTP service running and a user visiting a malicious site; it does not mean every installation has been accessed.
Complete briefing
Vulnerabilities & patches
3CVE-2026-123890: Critical Authentication Bypass in VMware vCenter
Published:
The most severe of these, CVE-2026-123890, is an authentication bypass using an alternate path or channel (CWE-288) in the VMware Directory Service component of vCenter. Broadcom assigned the vulnerability a critical CVSSv3 score of 9.8.
CVE-2026-123890: Critical Authentication Bypass in VMware vCenter
Published:
The most severe of these, CVE-2026-123890, is an authentication bypass using an alternate path or channel (CWE-288) in the VMware Directory Service component of vCenter. Broadcom assigned the vulnerability a critical CVSSv3 score of 9.8.
Sources · 8
DHI begins publishing Java4Shell and NGINX HTTP/2 fixed images
Published:
OpenJDK and Oracle disclosed Java4Shell (CVE-2026-114938) on November 17, and NGINX released fixes for two HTTP/2 flaws the next day. Docker began publishing fixed images for both in the Docker Hardened Images (DHI) catalog starting that same day.
DHI begins publishing Java4Shell and NGINX HTTP/2 fixed images
Published:
OpenJDK and Oracle disclosed Java4Shell (CVE-2026-114938) on November 17, and NGINX released fixes for two HTTP/2 flaws the next day. Docker began publishing fixed images for both in the Docker Hardened Images (DHI) catalog starting that same day.
Sources · 4
AI unearths ‘internet breaker’ flaws in Nginx web server
Published:
Nginx, the web server estimated to power a third of the internet, has shipped fixes for multiple vulnerabilities, including a critical unauthenticated remote code execution flaw that puts millions of websites at risk and has existed in the code for over a decade. Researchers from several security firms independently found some of the same flaws by using AI agents, adding to a string of recent AI-discovered vulnerabilities in widely used internet technologies such as PHP, Node, OpenSSH, and Java.
AI unearths ‘internet breaker’ flaws in Nginx web server
Published:
Nginx, the web server estimated to power a third of the internet, has shipped fixes for multiple vulnerabilities, including a critical unauthenticated remote code execution flaw that puts millions of websites at risk and has existed in the code for over a decade. Researchers from several security firms independently found some of the same flaws by using AI agents, adding to a string of recent AI-discovered vulnerabilities in widely used internet technologies such as PHP, Node, OpenSSH, and Java.
Sources · 8
Exploitation & attacks
4ARMORY CASTLE Exploits vSphere | Google Cloud Blog
Published:
On Oct. 15, 2026, Mandiant and Google Threat Intelligence Group (GTIG) published a blog post detailing recent activity in which the suspected PRC-nexus threat actor ARMORY CASTLE (formerly tracked as UNC5221) exploited CVE-2026-102764, a zero-day vulnerability in BeyondTrust Remote Support, which was patched in BT26-05. We also described some of the threat actor’s post-exploitation operations.
ARMORY CASTLE Exploits vSphere | Google Cloud Blog
Published:
On Oct. 15, 2026, Mandiant and Google Threat Intelligence Group (GTIG) published a blog post detailing recent activity in which the suspected PRC-nexus threat actor ARMORY CASTLE (formerly tracked as UNC5221) exploited CVE-2026-102764, a zero-day vulnerability in BeyondTrust Remote Support, which was patched in BT26-05. We also described some of the threat actor’s post-exploitation operations.
Internet traffic and attacks during the 2026 U.S. midterm elections | Cloudflare Blog
Published:
Internet traffic and attacks during the 2026 U.S. midterm elections
Internet traffic and attacks during the 2026 U.S. midterm elections | Cloudflare Blog
Published:
Internet traffic and attacks during the 2026 U.S. midterm elections
Your AWS Security, Identity, and Governance guide to re:Invent 2026 | AWS Security Blog
Published:
The AWS Security, Identity, and Governance team invites you to join us at re:Invent in Las Vegas from November 30 to December 4, 2026. You know all too well that AI agents now act on their own across the systems you run, and that changes what identity, detection, network protection, and governance have to cover.
Your AWS Security, Identity, and Governance guide to re:Invent 2026 | AWS Security Blog
Published:
The AWS Security, Identity, and Governance team invites you to join us at re:Invent in Las Vegas from November 30 to December 4, 2026. You know all too well that AI agents now act on their own across the systems you run, and that changes what identity, detection, network protection, and governance have to cover.
Axon: Download of Code Without Integrity Check in axon-mcp
Published:
axon rtk-install downloads a pre-built rtkx binary from GitHub releases and executes it via subprocess.run() on every text-compression call, without verifying its integrity — no checksum, signature, or attestation check anywhere in the download/install path. If the release pipeline or a maintainer token for sammyjdev/rtkx were ever compromised, every axon user who installs or auto-bootstraps rtkx would silently download and execute an unverified binary with the same privileges as the axon process.
Axon: Download of Code Without Integrity Check in axon-mcp
Published:
axon rtk-install downloads a pre-built rtkx binary from GitHub releases and executes it via subprocess.run() on every text-compression call, without verifying its integrity — no checksum, signature, or attestation check anywhere in the download/install path. If the release pipeline or a maintainer token for sammyjdev/rtkx were ever compromised, every axon user who installs or auto-bootstraps rtkx would silently download and execute an unverified binary with the same privileges as the axon process.
Malware
2Spying on Governments with Zero Lines of Malware: A Custom SoftEther Bridge in Webworm’s Toolkit
Published:
For a threat actor, the perfect malware is one they don’t have to write, and already does all they need. One example of this is shown by the China-aligned APT known as Webworm, which has been known to deploy the open-source SoftEther VPN into the networks of compromised governmental organizations to maintain low-profile, long-term access.
Spying on Governments with Zero Lines of Malware: A Custom SoftEther Bridge in Webworm’s Toolkit
Published:
For a threat actor, the perfect malware is one they don’t have to write, and already does all they need. One example of this is shown by the China-aligned APT known as Webworm, which has been known to deploy the open-source SoftEther VPN into the networks of compromised governmental organizations to maintain low-profile, long-term access.
Microsoft's official Agent Framework Python package on PyPI hijacked to push login-stealing malware | TechRadar
Published:
TechRadar reports that attackers compromised Microsoft’s official Agent Framework Python package on PyPI to distribute infostealing malware. The attack was detected and stopped quickly, but victims may still exist; the report says hundreds of thousands of developers rely on it daily.
Microsoft's official Agent Framework Python package on PyPI hijacked to push login-stealing malware | TechRadar
Published:
TechRadar reports that attackers compromised Microsoft’s official Agent Framework Python package on PyPI to distribute infostealing malware. The attack was detected and stopped quickly, but victims may still exist; the report says hundreds of thousands of developers rely on it daily.
PKI & cryptography
2OpenID4VP: X.509 chain validation accepts a forged certificate that reuses a trusted anchor's Subject DN without verifying its signature
Published:
A certificate whose Subject DN string matched a trusted anchor's Subject DN was accepted as the chain terminus **with no signature verification** against the anchor's public key.
OpenID4VP: X.509 chain validation accepts a forged certificate that reuses a trusted anchor's Subject DN without verifying its signature
Published:
A certificate whose Subject DN string matched a trusted anchor's Subject DN was accepted as the chain terminus **with no signature verification** against the anchor's public key.
gosaml2: Unauthenticated panic decrypting an EncryptedAssertion with a short or empty CipherValue
Published:
An attacker who can supply a crafted SAML Response for parsing can crash a `gosaml2` service provider by sending an `EncryptedAssertion` with AES-GCM encryption metadata and a `CipherValue` shorter than the GCM nonce size (12 bytes). The library panics with `slice bounds out of range` before any validation completes.
gosaml2: Unauthenticated panic decrypting an EncryptedAssertion with a short or empty CipherValue
Published:
An attacker who can supply a crafted SAML Response for parsing can crash a `gosaml2` service provider by sending an `EncryptedAssertion` with AES-GCM encryption metadata and a `CipherValue` shorter than the GCM nonce size (12 bytes). The library panics with `slice bounds out of range` before any validation completes.
Research & tools
2How AWS Security Incident Response triages your security findings | AWS Security Blog
Published:
AWS says its Security Incident Response service continuously triages threat findings from GuardDuty and Security Hub CSPM, but most of that work does not appear as cases. Case counts alone therefore do not show all triage activity.
How AWS Security Incident Response triages your security findings | AWS Security Blog
Published:
AWS says its Security Incident Response service continuously triages threat findings from GuardDuty and Security Hub CSPM, but most of that work does not appear as cases. Case counts alone therefore do not show all triage activity.
Guard Core: Quadratic ReDoS in built-in detection pattern freezes the event loop with a single request
Published:
A polynomial (quadratic) regular-expression denial of service in guard-core's built-in penetration-detection patterns allows any unauthenticated remote client to freeze **every** endpoint of an application for seconds per request, using only the default configuration. The triggering request completes without any detection match — no ban counter increments, no rate-limit violation is recorded — so it can be repeated indefinitely.
Guard Core: Quadratic ReDoS in built-in detection pattern freezes the event loop with a single request
Published:
A polynomial (quadratic) regular-expression denial of service in guard-core's built-in penetration-detection patterns allows any unauthenticated remote client to freeze **every** endpoint of an application for seconds per request, using only the default configuration. The triggering request completes without any detection match — no ban counter increments, no rate-limit violation is recorded — so it can be repeated indefinitely.
Identity & phishing
2Model Context Protocol Python SDK: OAuth client could send credentials to an authorization server chosen by the MCP server
Published:
A GitHub advisory says affected MCP Python SDK OAuth clients do not always validate the authorization server or bind credentials to it, potentially sending credentials to an endpoint chosen by a malicious server.
Model Context Protocol Python SDK: OAuth client could send credentials to an authorization server chosen by the MCP server
Published:
A GitHub advisory says affected MCP Python SDK OAuth clients do not always validate the authorization server or bind credentials to it, potentially sending credentials to an endpoint chosen by a malicious server.
k8s-mcp-server: Missing authentication on default SSE/HTTP transport allows unauthenticated Kubernetes tool execution
Published:
endpoint is dispatched with no authentication, session token, or Origin check.
k8s-mcp-server: Missing authentication on default SSE/HTTP transport allows unauthenticated Kubernetes tool execution
Published:
endpoint is dispatched with no authentication, session token, or Origin check.
Supply chain
1Credential stealer slipped into Microsoft's official Python agent framework - Help Net Security
Published:
Two releases of Microsoft’s official agent-framework package on the Python Package Index (PyPI) carried code that steals cloud and CI/CD credentials from developers’ systems. Agent Framework is Microsoft’s open-source toolkit for building AI agents and running them in workflows.
Credential stealer slipped into Microsoft's official Python agent framework - Help Net Security
Published:
Two releases of Microsoft’s official agent-framework package on the Python Package Index (PyPI) carried code that steals cloud and CI/CD credentials from developers’ systems. Agent Framework is Microsoft’s open-source toolkit for building AI agents and running them in workflows.
Cloud & container security
1Russian spies turn one cloud breach into a stepping stone for the next - Help Net Security
Published:
A Russian cyberespionage group is posing as employees of organizations it has already breached to phish new targets, Google Threat Intelligence Group (GTIG) found.
Russian spies turn one cloud breach into a stepping stone for the next - Help Net Security
Published:
A Russian cyberespionage group is posing as employees of organizations it has already breached to phish new targets, Google Threat Intelligence Group (GTIG) found.