This week’s stories share a common thread: trust that is inherited, automated, or assumed. Microsoft says attackers abused trusted connections from an IT management vendor to target more than 2,250 Microsoft 365 tenants at machine speed. A researcher documented one person running AI agents against more than 1,200 websites. AI also surfaced decades-old PHP flaws that attackers are now probing. Closer to home, Missouri state government shared what has worked in its AI pilots, and a Kansas City area credit union disclosed a phishing breach tied to a legacy email account. A global banking outage rounds out the week as a reminder of how much depends on a single cloud provider.
Microsoft says attackers used stolen ConnectWise app credentials and an AI tool to target 2,250+ Microsoft 365 tenants
Microsoft has attributed a large campaign against Microsoft 365 customers to Storm-0940, a group it assesses as China state-aligned and active since at least 2021. According to Microsoft, the group compromised ConnectWise’s multi-tenant cloud backup application and then used the permissions customers had already granted to ConnectWise services. That access did not require an interactive user sign-in.
The Register reports that the group attempted to authenticate against more than 2,250 tenants, at times reaching 2,900 attempts per minute. Microsoft says the group used ARTEX, an open-source AI penetration-testing framework, to automate reconnaissance, credential attacks, and data theft, and that in some cases it moved from targeting to privileged cloud access in under three hours. Microsoft has not said how many organizations were compromised or whether data was stolen. It also states the activity did not result from a vulnerability in its own products.
ConnectWise took its Cloud Backup, SaaS Security, and Cloud Management services offline. CEO Manny Rivelo said the affected connections were contained and services were being restored gradually. The Register notes that cutting those connections may not remove attackers who already had a foothold. The attribution and the role of AI are Microsoft’s assessments and have not been independently confirmed.
Why it matters
- Third-party apps connected to Microsoft 365 can carry broad, standing permissions that bypass user sign-in protections such as MFA prompts.
- Many organizations rely on IT management and backup vendors without reviewing what those connections can access.
- Automated attacks compress the time defenders have to notice and respond.
What organizations should consider
- If you use ConnectWise cloud services, follow Microsoft’s published guidance and keep affected application connections disabled until ConnectWise provides restoration instructions.
- Preserve sign-in and audit logs before removing apps or credentials, so investigators can confirm what happened.
- Review all third-party applications authorized in your Microsoft 365 tenant, including unused ones, and confirm that each still needs the permissions it holds.
Original article: The Register Supporting source: Microsoft Security Blog
One hacker’s AI agents broke into more than 100 company websites, researcher says
Security researcher Snorre Fagerland of NorForge found exposed control panels for Cairn, an open-source AI agent platform, being used to attack websites. According to The Next Web, the campaign began September 25. Its largest cluster, a Chinese-language operation spanning 14 panels and more than 1,100 attack projects, mainly targeted online shops in the US, Canada, Japan, and Australia. Reuters reported that the agents attempted more than 1,200 sites and reached more than 100 companies.
NorForge estimated an average success rate of about 10 percent. Fagerland estimated roughly $5,000 in AI usage costs, or about $50 per successful break-in, though NorForge’s own report puts the average near $5 per task and the article notes considerable uncertainty. The exposed panels also leaked hundreds of credentials and API keys, and 11 percent of those checked were still valid.
Reuters named Deere, SharkNinja, and Cummins among the targets. Deere said it saw no sign that sensitive information was accessed. SharkNinja said a standalone product support site was hit and the threat was contained. Cummins said it was looking into a possible attack on a website run by one of its vendors. The Next Web has not independently verified the intrusions, and which AI models were used remains unconfirmed. Cairn’s author said the tool was not designed for hacking.
Why it matters
- Low per-attack costs make smaller organizations’ websites and online stores economical targets.
- Vendor-run sites, such as support portals, can expose a company’s name even when its core systems are untouched.
- Attackers who leave their own tools exposed show that leaked API keys and credentials remain a common weakness.
What organizations should consider
- Inventory public-facing websites, including those run by vendors on your behalf, and confirm who is responsible for patching and monitoring each one.
- Check that API keys and credentials are not stored in exposed locations, and rotate any that may have been shared widely.
Original article: The Next Web
Missouri’s AI director shares early results and lessons from state agency pilots
Missouri state government has been testing AI tools under an executive order Gov. Mike Kehoe signed in January, which calls for a framework prioritizing efficiency, service quality, data privacy and security, and human oversight. In an interview with Route Fifty, Tim Marczewski, the state’s director of AI and innovation, said pilots began with about 100 employees across 17 agencies. Staff had to bring a specific use case to participate.
One early result came from the Missouri Agricultural and Small Business Development Authority. Its Beef and Dairy Producer Grants usually draw about 100 applicants, but this cycle drew 3,000. “They thought it was going to take from January to June to process applications, and we cut it down to one month,” Marczewski said, adding, “We saved about $200,000.” The work used AI with human review. These figures are the state’s own.
Marczewski said very large trainings did not work, and smaller, role-specific sessions such as lunch-and-learns were more effective. He cautioned against forcing AI on staff. The state surveys employees, using AI to analyze free-text responses, to judge whether tools actually reduce workload. For 2027, it is exploring AI agents to speed a technical review process that has taken six to 18 months.
Why it matters
- Missouri organizations now have a home-state example of AI adoption that pairs automation with human review.
- The lessons on training and measurement apply to any employer rolling out AI tools.
What organizations should consider
- Require a defined use case before giving staff access to new AI tools.
- Favor small, role-specific training over large one-time sessions.
- Measure whether AI reduces employee burden, not only whether tasks finish faster.
Original article: Route Fifty
Missouri Central Credit Union discloses phishing breach of a legacy email account
Missouri Central Credit Union, which serves Lee’s Summit and the greater Kansas City area, has notified individuals of a breach involving its email accounts, according to Teiss. The credit union traced the incident to suspicious activity in a legacy email account that belonged to Central Communications Credit Union. A forensic investigation found that an unauthorized person used a phishing scheme to access certain emails.
On May 14, 2026, the credit union finished reviewing the affected emails and confirmed they contained personal information of some individuals. Its notice was filed with Massachusetts regulators on October 12. Teiss reports that the information may include names, Social Security numbers, financial account information, and credit or debit card numbers, though the sample notice filed in Massachusetts left the specific data elements blank. The filing does not say how many people were affected.
The credit union says it is not aware of any fraud or identity theft connected to the incident. It is offering affected individuals a free Experian IdentityWorks membership and a support hotline.
Why it matters
- Accounts inherited through a merger or acquisition can remain active and overlooked long after the transition.
- Email inboxes often hold sensitive customer data that was never meant to be stored there.
- Phishing continues to be a direct route to regulated financial information.
What organizations should consider
- After a merger or system change, identify and retire or secure legacy email accounts.
- Review what sensitive information is routinely sent by email, and limit how long it stays in mailboxes.
- Members who receive a notice can enroll in the offered monitoring and watch account statements for unfamiliar activity.
Original article: Teiss
Critical PHP flaws found with AI tools are now being targeted
The PHP security team released patches on October 21 for two critical remote code execution flaws, CVE-2026-109618 and CVE-2026-109619, which have been nicknamed PHP2Shell. TechCrunch reports the flaws have existed in PHP versions since PHP 5 and were found by researchers using AI tools, including work tied to OpenAI and Google’s Wiz team.
According to BleepingComputer, Amazon Threat Intelligence observed the first exploitation attempts against its honeypot on October 17, four days before the patches, and said the flaws are “now being broadly exploited in the wild.” Reports differ on the impact so far. The PHP team said proof-of-concept code published before the patches did not work, and Dataminr reported activity limited to scanning and non-working exploits, with no known compromises. VirusTotal tracks more than 5.1 million exposed PHP hosts.
The UK’s NCSC asked organizations to update PHP servers immediately or take them offline until they can be patched, and TechCrunch reports the NSA gave similar guidance. WordPress has said the issue is not specific to WordPress, but sites running it need patched servers.
Why it matters
- PHP powers many business websites, including WordPress sites, often on hosting managed by a third party.
- Amazon warned that defenders should not assume the public disclosure date is the earliest date of exploitation.
What organizations should consider
- Confirm with your web host or developer that PHP has been updated with the October 21 patches.
- Identify any sites running unsupported PHP versions, which TechCrunch notes includes PHP 8.1 and older.
- If a server cannot be patched promptly, consider taking it offline, as the NCSC advises.
Original article: BleepingComputer Supporting source: TechCrunch
Revolut’s global outage traced to a Google Cloud disruption
Digital bank Revolut suffered a worldwide outage on Tuesday, October 20, that affected most of its services, BleepingComputer reports. Its status page listed 28 affected services, including sign-ins, card payments, transfers, currency exchange, stock trading, and live support chat. Thousands of users reported problems.
Reuters linked the outage to disruptions at Google Cloud’s London europe-west2 region following a “power event.” Revolut later said, “Due to a recent Google Cloud outage, customers may have experienced temporary service disruption on the Revolut app today,” and confirmed services were restored. The company said customer money and data remained safe throughout.
Why it matters
- A problem in one cloud region was enough to interrupt a global bank’s core customer services.
- Many organizations depend on SaaS vendors that, in turn, rely on a single cloud provider or region.
What organizations should consider
- Ask critical software and service vendors how they handle a regional cloud outage and how quickly they can recover.
- Identify which daily operations would stop if one key vendor went offline, and document a manual fallback.
- Keep customer communication templates ready so you can explain an outage quickly, as Revolut did.
Original article: BleepingComputer