Last week was a rough one for security, and AI was in the middle of most of it. For a regional carrier or a company that leans on an MSP, the four stories below are the ones I'd want on your desk, along with what I'd do about each. If you want a second set of eyes on any of these, talk with the Spej team.
An AI-driven attack got into Microsoft 365 tenants through a ConnectWise backup app.
On October 21 Microsoft attributed a compromise of ConnectWise's Cloud Backup service to a group it tracks as Storm-0940, which used an open-source agentic testing framework to reach customers' Microsoft 365 tenants through an app those customers had already approved. CyberScoop reported Microsoft's figure of more than 2,250 tenants at 2,900 sign-in attempts per minute, Microsoft said a successful sign-in does not necessarily mean data was taken, and ConnectWise said shutting off the app does not remove access already set up inside a customer tenant. Ask your MSP for every third-party app with consent into your tenant and what each one is allowed to do.
PHP2Shell turned into a patch scramble, and it is now being exploited.
On October 21 the PHP security team shipped emergency fixes for two critical flaws, nicknamed PHP2Shell, that let an attacker run code on a web server without logging in, and CISA told organizations to patch immediately or take unpatched PHP systems offline, including anything still on unsupported PHP 8.1 or older. BleepingComputer then reported the first confirmed exploitation in the wild. Ask for an inventory of every system running PHP, including vendor-hosted sites, and written confirmation that each one is patched and restarted.
Banks are blocking the AI agents their customers let into their accounts.
Axios reported on October 22 that major U.S. banks are detecting third-party AI agents inside customer accounts, generally because customers handed over usernames and passwords, and that Morgan Stanley has started blocking those agents while JPMorgan, Bank of America, and Citi are restricting them, with most banks moving toward read-only access. Write down what an agent may do on a customer portal, because checking a status is different from changing a payee or bank details, and say plainly that an employee giving an agent their work login is unsanctioned AI use.
The NAIC's new AI exam questions were open for comment until October 30.
On October 16 the NAIC exposed version 6.0 of its AI Risk Evaluation Supplement, the regulator-facing questionnaire 12 states piloted this year, with comments due Friday, October 30. InsureAI Wire's review of the draft said it asks for an AI system inventory, whether a use case is agentic and what it can do without human approval, and how a company finds unsanctioned AI use. A regional carrier should start that inventory now, including AI inside vendor tools and what staff use on their own.
First published on Spej's Substack.
