Vulnerability discovery and disclosure is accelerating. The CVE Program published 48,244 CVE Records in 2025 and 35,872 in the first half of 2026 alone, about 74% of last year's total (CVE Program metrics). Last week brought a surge of critical vulnerabilities and compromises: two critical PHP remote code execution flaws (CVSS 9.2, patched October 21) whose details leaked before the fix, eMudhra, a certificate authority whose weak intermediate keys could be recovered, a malicious LiteLLM release, an Opencode security incident, and a critical patch for a CVSS 10.0 unauthenticated remote code execution flaw in GitLab. Three things to check on your team:
1. Make patching boring# #
Could you answer "where do we run PHP, which version, and how many copies of our certificate trust store exist?" in minutes? The CA incident meant updating trust stores, containers, and language runtimes, not just browsers. A fast, tested build-and-deploy pipeline turns a week of back-to-back CVE patching into routine work, so engineers can keep shipping features. Inventory your versions, then rehearse a fleet-wide update. Exercise remediation at each stage of the pipeline: ensure your SBOMs are up-to-date, that actionable supply chain alerts are separated from noise, and that version-bumps can be automated as much as possible.
2. Put real gates on releases# #
In the LiteLLM incident, attackers bypassed a two-person approval system by stealing tokens from two engineers. Multi-party approval systems only work when used with strong authentication methods and aren't satisfiable by secrets held in CI jobs or reachable by admins. Consider gating approval behind authentication methods such as OIDC instead of relying on static secrets. Overridable branch protections provides a source of alert, but doesn't prevent the incident. Teams should maintain and practice a plan for revoking tainted releases. Audit who can approve and bypass, and alert on both.
3. Be able to rotate any key, fast# #
As industry reliance on AI agents and gateways holding privileged credentials grows, a leaked key goes further than it used to. The LiteLLM chain started with a master key committed to a public repo that was never revoked. In October it was used to extract a GitHub org owner's access token out of an experimental agent platform. Deleting a leaking commit isn't remediation; revoke, rotate, and verify the repair of each compromised secrets. Keep human admin tokens out of agents and CI in favor of scoped, short-lived credentials. Confirm every past secret exposure ended in a verified rotation, and assume any leaked keys to be held by attackers.
In Conclusion# #
Improperly configured and overprivileged identities increase the blast radius of exploitation. Broad usage of keys and credentials makes life easier for attackers attempting to pivot across your network. As the rate of vulnerability discovery surges, identity continues to be the most important boundary to securing your enterprise. In the post-ai world, proper authentication and authorization of every action in your workflow can be the difference between a widespread compromise and a benign-positive security alert.