CVSS 10 in PraisonAI's Claude workflow, plus two Cisco IOS XE crashers
A fork-based command injection lets any outside contributor run code in PraisonAI's GitHub Actions runners (CVE-2026-48168, CVSS 10). Two unauthenticated Cisco IOS XE DoS bugs (CVSS 8.6 each) can reload your routers with a single packet. Keycloak SAML bypass and a GStreamer memory bomb round out the day.
A perfect 10 just dropped. PraisonAI's bundled Claude GitHub Actions workflow lets any outside contributor inject shell commands through a branch name, giving them code execution on the Actions runner with write access to your repo. No exploits spotted in the wild yet, but the bar to pull this off is absurdly low. Behind that, two unauthenticated Cisco IOS XE denial-of-service bugs and a Keycloak SAML auth bypass round out a day you don't want to ignore.
Today's CVEs
Sorted by urgencyCVE-2026-48168
NVDAn attacker can open a pull request against any PraisonAI repo that uses the bundled Claude GitHub Actions workflow, stuff shell commands into the branch name, then comment @claude to trigger arbitrary code execution inside the Actions runner. Because the runner holds a GitHub App token with write permissions and OIDC access, the attacker can push code to the repo, manipulate issues and PRs, or steal OIDC tokens. No special access is needed: any outside contributor on a fork can pull this off.
- Affected estate
- Anyone running PraisonAI prior to 4.6.40 with the bundled Claude GitHub Actions workflow enabled on their repos
- How to check
- Search your .github/workflows/ directory for the Claude workflow file. Check your PraisonAI version (pip show praisonai). If the workflow exists and the version is below 4.6.40, you're exposed.
- Included because
- unauthenticated; internet-facing; CVSS 10.0; command injection in CI/CD pipeline; no interaction required beyond a comment
- Action
- Upgrade PraisonAI to 4.6.40+. Review recent Actions logs for fork-originated PRs with shell metacharacters in branch names.
- Why it matters
- CVSS 10.0: any anonymous GitHub user can execute arbitrary code with write access to your repository and steal OIDC tokens.
- Source
- PraisonAI GitHub advisory
Evidence trail
- NVD: View source
CVE-2026-20301
NVDA remote, unauthenticated attacker can send a single malformed XMCP packet to a Cisco IOS or IOS XE device and crash it, forcing an unexpected reload. No credentials or knowledge of the XMCP client username is required. If you have XMCP (External Client protocol) enabled on any routers or switches, they're sitting ducks for a DoS.
- Affected estate
- Network engineers running Cisco IOS or IOS XE devices with XMCP enabled
- How to check
- Run 'show running-config | include xmcp' on your devices. If XMCP is configured, you're exposed. Also check your IOS/IOS XE version against Cisco's advisory.
- Included because
- unauthenticated; remote; network infrastructure; CVSS 8.6; device reload DoS
- Action
- Apply the fixed IOS/IOS XE release from Cisco. As a workaround, disable XMCP or apply an ACL to block untrusted XMCP traffic.
- Why it matters
- An unauthenticated attacker can remotely crash your network devices, causing outages across your infrastructure.
- Source
- Cisco Security Advisory
Evidence trail
- NVD: View source
CVE-2026-20263
NVDA crafted BEEP SOAP request sent to a Cisco IOS XE device can crash it and force a reload, no authentication required. This is another unauthenticated remote DoS in IOS XE's protocol handling. If BEEP is reachable from untrusted networks, you need to act fast.
- Affected estate
- Network engineers running Cisco IOS XE devices with the BEEP feature exposed
- How to check
- Review your running config for BEEP-related configuration. Check your IOS XE version against Cisco's advisory for affected releases.
- Included because
- unauthenticated; remote; network infrastructure; CVSS 8.6; device reload DoS
- Action
- Upgrade to the patched IOS XE release. As a stopgap, apply ACLs to block untrusted access to BEEP services.
- Why it matters
- An unauthenticated attacker can remotely crash your switches or routers, causing network outages.
- Source
- Cisco Security Advisory
Evidence trail
- NVD: View source
CVE-2026-18649
NVDThe H.264 and H.265 RTP depayloaders in GStreamer's gst-plugins-good package never cap the size of their reassembly buffer. A remote attacker can send a stream of RTP fragments that never finish, causing memory usage to grow until the process dies. Any application using GStreamer to receive RTP video streams (media servers, video conferencing apps, surveillance systems) is vulnerable.
- Affected estate
- Anyone running applications that use GStreamer's gst-plugins-good for RTP video processing, especially media servers, streaming platforms, and video conferencing systems on Linux
- How to check
- Run 'gst-inspect-1.0 rtph264depay' to confirm the plugin is installed, then check 'dpkg -l gstreamer1.0-plugins-good' or 'rpm -q gstreamer1-plugins-good' for the installed version.
- Included because
- unauthenticated; remote; common multimedia framework; CVSS 7.5; denial of service
- Action
- Update gst-plugins-good via your package manager or build from the latest upstream release that includes the fix.
- Why it matters
- An unauthenticated remote attacker can exhaust process memory and kill any GStreamer-based media application by sending crafted RTP fragments.
- Source
- Red Hat / GStreamer upstream advisory
Evidence trail
- NVD: View source
CVE-2026-16443
NVDWhen Keycloak imports SAML identity provider metadata that doesn't include specific key usage attributes, it silently disables signature validation on SAML responses, even when a signing certificate is present. An unauthenticated attacker who knows a user's external identifier can forge a SAML response and log in as that user. This is a serious authentication bypass for anyone using SAML identity brokering in Keycloak.
- Affected estate
- Teams running Red Hat Build of Keycloak (or upstream Keycloak) with SAML identity brokering configured
- How to check
- In the Keycloak admin console, go to Identity Providers, select your SAML IdP, and check whether 'Validate Signatures' is actually enforced. Review your imported IdP metadata XML for missing 'use' attributes on KeyDescriptor elements.
- Included because
- unauthenticated; authentication bypass; internet-facing SSO; CVSS 7.4; SAML signature validation disabled silently
- Action
- Upgrade Keycloak to the patched release. After upgrading, re-import your SAML IdP metadata and confirm signature validation is active.
- Why it matters
- An attacker can forge SAML responses and take over user accounts without any credentials, bypassing your SSO entirely.
- Source
- Red Hat Security Advisory
Evidence trail
- NVD: View source
One email, every Wednesday morning.
SubscribeFrom the field notes
From this beat
Read the rest of the field notes โ