CYBER THREATCAST

Upgrade exposed PHP installations to fixed releases immediately, prioritizing Apache mod_php, Drupal, Joomla, and unsupported PHP deployments; isolate systems that cannot be upgraded. Review them for compromise because CVE-2026-109618 and CVE-2026-109619 have enabled remote code execution in the wild, public exploits are available, and attackers are modifying payloads to evade WAFs. Where the configuration for unpatched CVE-2026-115560 is present, temporarily disable session upload progress pending a fix.

Remove LiteLLM 1.107.0 and affected container tags from every environment, investigate systems that installed or pulled them, and rotate credentials those hosts could access. All seven 1.107.0 PyPI wheels reportedly contained credential-stealing code, and the affected tags include latest, main-latest, main-stable, and v1.107.0. Investigators also observed attempted access to AWS-hosted AI models, so include relevant cloud credentials and AI-service access in the review.

Apply Arista’s fixes for CVE-2026-114625 to every on-premises VeloCloud Orchestrator and review basic-user accounts for suspicious elevation or operator-level activity. Arista disclosed active exploitation of this authenticated SQL-injection flaw, which can turn a basic account into operator access, and CISA added it to the Known Exploited Vulnerabilities catalog.

Verify any Keystone hardware wallet received after October 20 through the additional authentication process advised for old-key devices before trusting it with funds. Keystone’s stolen device-authentication key could allow counterfeit devices to pass existing authenticity checks, while exposed order data covered 25,964 completed purchases and 12,873 abandoned checkouts. Keystone customers should also treat communications using their leaked names, contact details, shipping addresses, products, prices, or order records as potentially fraudulent.

Inspect browser-writable directories, especially Downloads, for untrusted AGENTS.md or CLAUDE.md files before launching OpenAI Codex or Anthropic Claude Code, and review agent configuration for persistent server changes. If an agent processed a suspicious instruction file, investigate executed commands and rotate exposed session or API tokens. Researchers reproduced malware execution and token exfiltration through these files, and Codex followed the malicious instructions in more than 90% of tests.