← All briefings

DB-GPT · piflow · Cognition

AI-generated briefing. Verify critical findings against the linked primary sources.

Prepared for
  • SOC Analysts
  • Incident Responders
  • Vulnerability Management
  • Security Leadership

Executive Summary

Newly reported vulnerabilities in AI-agent and data-science platforms include unauthenticated remote code execution in DB-GPT, piflow and Cognition, as well as unauthenticated API access in Letta. These systems warrant urgent exposure checks, particularly where deployed with access to host tools or sensitive data. Separately, Rapid7 reports active intrusions exploiting two NetScaler vulnerabilities, while ransomware operators are exploiting PAN-OS CVE-2026-0257.

Critical Vulnerabilities

CVE-2026-77932 — DB-GPT

  • Severity: CVSS 9.3.
  • EPSS: 0 — no elevated probability indicated by the supplied score.
  • Technical detail: DB-GPT through 0.8.2 reportedly has unauthenticated remote code execution through /api/v1/chat/react-agent. The endpoint uses a mock identity function that grants callers access; exposure depends on whether the service is reachable by an attacker.
  • Exploitation status: No confirmed exploitation reported in the supplied data.
  • Remediation: Check for a vendor-fixed release and apply it. Restrict access to DB-GPT interfaces and review host and application activity for unexpected commands or changes.

CVE-2026-109442 — piflow

  • Severity: CVSS 9.3.
  • EPSS: 0 — no elevated probability indicated by the supplied score.
  • Technical detail: piflow 2.0.0–2.0.1 reportedly exposes unauthenticated /chat and /chat/stream endpoints that can direct its DeepAgents chat agent to execute actions. Risk is highest where the service is internet-accessible or the agent has access to sensitive tools, files or systems.
  • Exploitation status: No confirmed exploitation reported in the supplied data.
  • Remediation: Check for and deploy a fixed release. Until then, restrict endpoint access and disable or limit agent capabilities that can interact with the host or other systems.

CVE-2026-109437 — Cognition

  • Severity: CVSS 9.2.
  • EPSS: 0 — no elevated probability indicated by the supplied score.
  • Technical detail: Cognition versions before 0.13.0 reportedly default to local execution for the agent sandbox and provide no authentication on REST/SSE APIs. An attacker able to reach those interfaces may be able to execute code on the host.
  • Exploitation status: No confirmed exploitation reported in the supplied data.
  • Remediation: Upgrade to 0.13.0 or later. Restrict API exposure and verify that agent execution is isolated from production systems and credentials.

CVE-2026-77936 — Letta

  • Severity: CVSS 9.3.
  • EPSS: 0 — no elevated probability indicated by the supplied score.
  • Technical detail: Letta through 0.16.8 reportedly ships a Compose configuration that does not enable secure mode, leaving API routes unauthenticated and exposing a superuser PostgreSQL credential. The described risk is most relevant to deployments using the affected default configuration; verify actual configuration and network reachability.
  • Exploitation status: No confirmed exploitation reported in the supplied data.
  • Remediation: Check for a fixed release and secure the deployment immediately. Enable authentication, change exposed database credentials, and restrict access to API and database ports.

CVE-2026-53995 — MLflow

  • Severity: CVSS 8.8.
  • EPSS: 0 — no elevated probability indicated by the supplied score.
  • Technical detail: MLflow through 3.17.1 reportedly has a path-traversal flaw in the FileStore tracking backend. Unauthenticated attackers may control filesystem paths by submitting a crafted trace ID; impact depends on the service’s accessible filesystem and deployment permissions.
  • Exploitation status: No confirmed exploitation reported in the supplied data.
  • Remediation: Check for a fixed release and apply it. Restrict access to tracking services and assess whether the FileStore contains sensitive or writable paths.

ONGOING

  • CVE-2026-108983 (GeoVision GV-ASManager): Previously reported; no new status supplied. Check vendor guidance and restrict management interfaces.
  • CVE-2026-111998 (Vanus): Previously reported; no new status supplied. Upgrade when fixed release is available and restrict gateway access.
  • CVE-2026-112005 (WhoDB): Previously reported; no new status supplied. Upgrade to 0.135.0 or later.
  • CVE-2026-112051 (Xinference): Previously reported; no new status supplied. Check for a fix and restrict endpoint access.
  • CVE-2026-112001 (ProjectSend): Previously reported; no new status supplied. Upgrade to 2.7.0 or later.
  • CVE-2026-102764 (BeyondTrust Remote Support): Active exploitation previously reported; patch and investigate affected appliances.
  • CVE-2025-61624 (Fortinet): Active exploitation previously reported; patch affected products and investigate for PEERSTITCH indicators.

European Advisories

  • BSI issued a critical advisory for Collabora Online Built-in CODE Server in Nextcloud: an unauthenticated remote attacker may execute code with web-server-user privileges. Review WID-SEC-2026-4011, identify affected deployments, and apply vendor guidance.
  • BSI updated its critical GNU libc advisory, warning of potential file manipulation, denial of service and other unspecified impacts: WID-SEC-2026-1190. Review the advisory for affected distributions and available updates.
  • BSI also updated its critical Ruby on Rails advisory, describing information disclosure that may contribute to further compromise: WID-SEC-2026-2574. Check the advisory for affected versions and remediation.
  • BSI updates include Linux kernel, Red Hat, Apache HTTP Server, Xen and other products. Review the linked CERT-Bund advisory feed for applicable updates; supplied records do not detail specific new fixes.

Active Threats and Campaigns

  • NetScaler intrusions: Rapid7 reports intrusions chaining CVE-2026-88771 and CVE-2026-88779 against affected Citrix NetScaler ADC appliances. Observed activity included root code execution, persistence, PHP web shells and internal tunnelling. Identify affected appliances, patch per vendor guidance, and investigate for persistence and unusual outbound or internal traffic. Rapid7 analysis
  • PAN-OS exploitation: Unit 42 reports ransomware operators identified as Spikey Scorpius (Qilin) and Tarnished Scorpius (Inc) exploiting CVE-2026-0257. Prioritize patching and review firewall telemetry for compromise. Unit 42 threat brief
  • Plex2Shell: Reporting says attackers are chaining Plex Media Server flaws to install cryptocurrency miners and Linux proxy software following public PoC release. Review exposed Plex systems and investigate for unexpected processes and outbound connections. The Hacker News

Security News and Context

  • Some Rank Math PRO WordPress plugin downloads or updates distributed during the reported October 15–16 window contained a backdoor. Treat sites that installed and activated affected copies as potentially compromised; BleepingComputer.
  • AOMEI’s update infrastructure was reportedly compromised to distribute malware through its backup software. Validate software integrity and investigate affected endpoints; BleepingComputer.
  1. Restrict access to DB-GPT, piflow, Cognition and Letta; identify affected versions/configurations and apply vendor fixes as available.
  2. Review MLflow FileStore deployments for exposure, apply fixes and check for unexpected filesystem changes.
  3. Investigate NetScaler and PAN-OS telemetry for exploitation, persistence and lateral movement; patch affected appliances urgently.
  4. Check Plex, Rank Math PRO and AOMEI deployments for exposure or compromise; isolate suspicious systems and preserve relevant evidence.
  5. Review BSI’s Collabora advisory and assess affected Nextcloud deployments.
  6. Continue remediation of previously reported items: BeyondTrust, Fortinet, GeoVision GV-ASManager, Vanus, WhoDB, Xinference and ProjectSend.