Security Digest: September 27, 2026 - 23 Critical Vulnerabilities

Today’s alert is severe: we have 23 critical and high-severity vulnerabilities across routers, sandboxing runtimes, confidential-computing platforms, and AI/automation tools. Several issues have public exploits or active proof-of-concepts, and multiple flaws enable remote code execution, sandbox escape, SSRF, credential theft, and unauthorized access.

· 15 min read

Executive Summary

Today’s alert is severe: we have 23 critical and high-severity vulnerabilities across routers, sandboxing runtimes, confidential-computing platforms, and AI/automation tools. Several issues have public exploits or active proof-of-concepts, and multiple flaws enable remote code execution, sandbox escape, SSRF, credential theft, and unauthorized access.

Act now: patch exposed systems first, isolate unpatched services, and treat any internet-facing instance of D-Link, Coolify, AzuraCast, vm2, MONAI, Contrast, or Heym as potentially compromised until verified.

Critical Vulnerabilities

  • CVE-2026-100740: D-Link DIR-895L tunnel parser out-of-bounds write
    • Impact: Remote attackers can trigger memory corruption; the exploit is public and may already be in use.
    • Affected Systems: D-Link DIR-895L A1_102b07
    • Immediate Action: Remove the device from direct internet exposure now; block WAN access to management and VPN/L2TP services.
    • Mitigation: Apply vendor firmware as soon as available; if no patch exists, replace or isolate the device behind strict firewall rules.
  • CVE-2026-100721: vm2 NodeVM external-module resolver sandbox escape
    • Impact: Untrusted code can escape the sandbox and execute arbitrary code in the host process.
    • Affected Systems: vm2 before 3.12.2
    • Immediate Action: Upgrade immediately and stop running untrusted guest code in affected environments.
    • Mitigation: Update to 3.12.2 or later; review any use of require.external with custom resolvers and context: 'host'.
  • CVE-2026-100856: AzuraCast remote relay password code injection
    • Impact: Authenticated users with RemoteRelays permission can execute arbitrary Liquidsoap code, steal API keys, or disrupt stations.
    • Affected Systems: AzuraCast before 0.23.6
    • Immediate Action: Restrict RemoteRelays access and review relay password fields for suspicious interpolation syntax.
    • Mitigation: Upgrade to 0.23.6 or later.
  • CVE-2026-100865: Heym workflow eval, webhook bypasses, OAuth abuse, plaintext tokens
    • Impact: Attackers can run arbitrary Python, trigger workflows without auth, steal OAuth codes, and recover bearer tokens from the database.
    • Affected Systems: Heym before 0.0.53
    • Immediate Action: Disable untrusted workflow edits/templates and rotate all workflow, portal, HITL, and OAuth secrets.
    • Mitigation: Upgrade to 0.0.53 or later; invalidate exposed tokens immediately.
  • CVE-2026-100864: heym expression engine sandbox escape
    • Impact: Authenticated users can execute arbitrary Python on the backend.
    • Affected Systems: heym before 0.0.91
    • Immediate Action: Suspend access to expression-editing features until patched.
    • Mitigation: Upgrade to 0.0.91 or later.
  • CVE-2026-100839: Contrast BadAML guest-kernel attack
    • Impact: A malicious host can execute code in the guest kernel and read or modify confidential guest memory.
    • Affected Systems: Contrast before 1.18.0 on AMD SEV-SNP Metal-QEMU-SNP and Metal-QEMU-SNP-GPU
    • Immediate Action: Prioritize upgrades for confidential workloads and limit trust in host-controlled ACPI inputs.
    • Mitigation: Upgrade to 1.18.0 or later.
  • CVE-2026-100844: MONAI nnUNetV2Runner command injection
    • Impact: Crafted YAML or CLI values can execute shell commands as the job user.
    • Affected Systems: MONAI before 1.6.0
    • Immediate Action: Stop processing untrusted training configs and quarantine suspicious job files.
    • Mitigation: Upgrade to 1.6.0 or later.
  • CVE-2026-100833: Contrast policy flaw allows image substitution
    • Impact: Attackers with Kata agent API access can swap container images and undermine workload integrity.
    • Affected Systems: Contrast 1.14.0 through 1.23.0
    • Immediate Action: Reissue runtime policies and verify image digest enforcement.
    • Mitigation: Upgrade to 1.23.1 or later.
  • CVE-2026-100838: Contrast CopyFile policy flaw enables guest root writes
    • Impact: A malicious host process can overwrite critical guest files and take over the workload.
    • Affected Systems: Contrast before 1.19.1
    • Immediate Action: Block untrusted VSOCK access and apply the rego policy fix if you cannot upgrade immediately.
    • Mitigation: Upgrade to 1.19.1 or later.
  • CVE-2026-100857: AzuraCast Liquidsoap string injection
    • Impact: Authenticated users with Media or Profile permissions can inject code that runs as the AzuraCast user.
    • Affected Systems: AzuraCast before 0.23.4
    • Immediate Action: Review playlist URLs and metadata for Liquidsoap interpolation payloads.
    • Mitigation: Upgrade to 0.23.4 or later.
  • CVE-2026-100843: MONAI unsafe pickle deserialization
    • Impact: Malicious pickle files can execute arbitrary system commands.
    • Affected Systems: MONAI before 1.6.0
    • Immediate Action: Treat all .pkl inputs as untrusted and block external model files.
    • Mitigation: Upgrade to 1.6.0 or later.
  • CVE-2026-100840: MONAI bundle engine arbitrary code execution
    • Impact: Malicious bundles can execute code when loaded or run.
    • Affected Systems: MONAI through 1.6.0
    • Immediate Action: Stop loading third-party bundles until patched.
    • Mitigation: Upgrade to 1.6.0 or later and restrict bundle sources.
  • CVE-2026-100850: AzuraCast AutoDJ remote URL SSRF and local file read
    • Impact: Users with Media permission can read local files and query internal endpoints; no patch was available at publication.
    • Affected Systems: AzuraCast before 0.23.8
    • Immediate Action: Disable remote playlist fetches and restrict Media permission immediately.
    • Mitigation: Apply vendor guidance and monitor for an official fix.
  • CVE-2026-100851: AzuraCast station profile leaks broadcast credentials
    • Impact: View-only users can retrieve Icecast/Shoutcast admin and source passwords.
    • Affected Systems: AzuraCast before 0.23.8
    • Immediate Action: Rotate all streaming credentials now.
    • Mitigation: Upgrade to 0.23.8 or later.
  • CVE-2026-100846: MONAI unsafe pickle deserialization
    • Impact: Attacker-supplied pickle files can trigger code execution during deserialization.
    • Affected Systems: MONAI before 1.5.2
    • Immediate Action: Block untrusted .pkl files and scan pipelines for deserialization paths.
    • Mitigation: Upgrade to 1.5.2 or later.
  • CVE-2026-100847: AzuraCast DQL injection in sortOrder
    • Impact: Attackers can extract credentials and station settings from the database.
    • Affected Systems: AzuraCast before 0.23.8
    • Immediate Action: Restrict access to list/search endpoints and inspect logs for malformed sort parameters.
    • Mitigation: Upgrade to 0.23.8 or later.
  • CVE-2026-100835: Contrast attestation relay attack
    • Impact: Attackers can impersonate a Coordinator or workload and defeat attested TLS identity checks.
    • Affected Systems: Contrast before 1.16.0
    • Immediate Action: Verify attestation flows and isolate any network paths that can be intercepted.
    • Mitigation: Upgrade to 1.16.0 or later.
  • CVE-2025-71423: Contrast workload secret exposed in logs
    • Impact: Workload secrets can be read from pod logs.
    • Affected Systems: Contrast 1.9.0 through 1.12.1
    • Immediate Action: Assume secrets are exposed if pods/logs were accessible.
    • Mitigation: Upgrade to 1.12.2 or later and rotate affected secrets.
  • CVE-2026-100744: Coolify route-level authorization bypass
    • Impact: Remote attackers can bypass authorization; exploit is public.
    • Affected Systems: Coolify up to 4.1.2
    • Immediate Action: Upgrade immediately and restrict access to the admin interface.
    • Mitigation: Upgrade to 4.2.0 or later.
  • CVE-2026-100746: Coolify GitHub App setup missing authentication
    • Impact: Remote attackers can abuse the GitHub redirect flow; exploit is public.
    • Affected Systems: Coolify up to 4.1.0
    • Immediate Action: Disable exposed GitHub app setup flows until patched.
    • Mitigation: Upgrade to 4.1.1 or later.
  • CVE-2025-71425: Contrast initializer logs workload secrets
    • Impact: Secrets may be exposed in Kubernetes logs and log storage.
    • Affected Systems: Contrast before 1.8.1, especially default log level installations
    • Immediate Action: Rotate workload secrets and review log access immediately.
    • Mitigation: Upgrade to 1.8.1 or later.
  • CVE-2026-100848: AzuraCast Remote Relay SSRF
    • Impact: Users with RemoteRelays permission can force repeated requests to internal, loopback, or metadata endpoints.
    • Affected Systems: AzuraCast before 0.23.8 and main branch as of 2026-07-31
    • Immediate Action: Restrict RemoteRelays permission and block access to metadata and private ranges.
    • Mitigation: No patched version was available at publication; apply vendor guidance.
  • CVE-2025-71426: Contrast rogue Coordinator recovery flaw
    • Impact: Attackers can impersonate a Coordinator and recover workload secrets if traffic is redirected.
    • Affected Systems: Contrast before 1.4.1
    • Immediate Action: Validate Coordinator certificates against a trusted reference before any recovery action.
    • Mitigation: Upgrade to 1.4.1 or later.

Previously Alerted

What to Do Now

  1. Patch internet-facing systems first: vm2, Coolify, AzuraCast, MONAI, Heym, and Contrast should be updated before the next maintenance window.
  2. Disable risky features: untrusted workflow editing, remote relay fetches, third-party bundles, custom sandbox resolvers, and exposed management APIs.
  3. Rotate credentials and tokens: streaming passwords, workload secrets, portal/HITL tokens, OAuth codes, and any secrets that may have been logged or exposed.
  4. Contain exposure: block WAN access to admin panels, restrict VSOCK/Kata agent paths, and prevent access to private/metadata IP ranges.
  5. Assume compromise where exploits are public: investigate D-Link DIR-895L, Coolify, and any externally reachable AzuraCast or vm2 deployments immediately.

Verification steps: confirm versions from package manifests, container images, and appliance firmware; check whether vulnerable endpoints are reachable from the internet; review logs for suspicious relay URLs, malformed sort parameters, unexpected workflow edits, and unusual deserialization or shell activity.

Monitoring recommendations: alert on new admin sessions, outbound requests to 169.254.169.254 and RFC1918 ranges, unexpected child processes from web or job runners, secret access from logs, and any changes to workflow, relay, or policy configuration.

Related Resources

  • Internal blog post: Pending — “Today’s Critical Vulnerabilities: Response Playbook”
  • Internal blog post: Pending — “How to Prioritize Public-Exploit CVEs in Production”
  • Official vendor advisories: D-Link, vm2, AzuraCast, Contrast, MONAI, Heym, and Coolify security advisories

Keep reading