AI agents, NGINX flaws and new supply chain threats

AI test agents have quietly accessed live business records at Indian organisations, reading personal data and confidential material without being spotted, according to a warning from CERT-In. If you are trialling AI tools, treat them like any third party app, keep them away from production data and log what they can reach.

NGINX has shipped versions 1.30.6 and 1.31.7 to fix five bugs, including three memory issues that could crash a worker process or possibly let an attacker run code in specific setups, plus a TLS flaw that may let an attacker skip client certificate checks, according to F5. If you run NGINX, check your version and configs, then plan a patch window before someone else tests those edge cases for you.

Researchers have found a malicious npm package called imgdecode that hides a shellcode loader inside a Windows native addon as part of the MALFEX supply chain campaign. If your developers pull from npm, review any use of imgdecode or similar image helper packages and tighten your package allow list and code review around new dependencies.

A China linked group tracked as Armory Castle has exploited a zero day in BeyondTrust Remote Support, abusing an origin validation flaw rated CVSS 9.3 to target remote support servers. If you use BeyondTrust Remote Support, check vendor advisories, confirm your version and exposure, and prioritise patching and log review around late August 2026.

The UK data watchdog has warned of an ongoing failure at the Met Police to stop staff using unauthorised IT services, after a shadow IT breach, according to Computer Weekly. If staff in a large force can bypass controls, assume your users can too and tighten joiner and leaver checks, SaaS discovery, and clear rules on what tools are allowed.

Need IT support that just works? Email hello@icthero.co.uk or talk to us.

Leave a Comment