The PHP2Shell scramble widens as governments pull websites offline and Amazon reports attacks before the fix, while Microsoft names Chinese hackers using AI agents behind the ConnectWise breach and a popular AI developer tool is poisoned.
A day after PHP shipped emergency fixes for the critical PHP2Shell flaws (CVE-2026-109618 and CVE-2026-109619), hundreds of websites were taken offline as a precaution, including government sites in Germany, France, Belgium, the Netherlands, Finland and the Czech Republic, and parts of Wikipedia. Amazon says one of its decoy systems saw an exploitation attempt on 17 October, four days before the public disclosure, and further attempts on 21 October from what looked like multiple attackers. The UK government says it is working to identify any exposure in its own online services, and the US National Security Agency warned the bugs can give attackers a direct path to a server's files or database.
Why it matters: Attackers were probing these flaws before most defenders could patch, and the PHP Foundation says a leaked video and an early release of fixed builds sped up exploit development. Only PHP 8.2 and newer get official fixes, and estimates suggest around half of PHP systems run older versions, so anyone running PHP should update to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 now, or take exposed systems offline until they can.
Microsoft says a China state-aligned group it tracks as Storm-0940 was behind the recent compromise of ConnectWise's Cloud Backup application. The group used ARTEX, an AI-driven penetration testing framework, to automate scanning, password spraying and abuse of the stolen app credentials, at one point attempting to sign in to more than 2,250 Microsoft 365 tenants at up to 2,900 attempts a minute. In some cases it went from first targeting to privileged cloud access in under three hours.
Why it matters: This shows attackers using AI agents to exploit one trusted supplier connection across many customers at machine speed, leaving defenders very little time to react. Organisations that used the affected ConnectWise apps should keep them disabled until ConnectWise says otherwise, and check for new app credentials or role changes, which Microsoft warns can act as a backdoor even after the original app is fixed.
Attackers hijacked the release process of LiteLLM, a widely used Python library for connecting apps to AI models, and published a malicious version 1.107.0 to PyPI and Docker on 21 October. The bad release was live for roughly two to three hours and installed a credential stealer plus a backdoor that runs whenever Python starts, even if LiteLLM is never used. LiteLLM says the attacker used stolen GitHub tokens from three engineers' accounts and built the release from a lookalike repository.
Why it matters: LiteLLM often sits on systems holding API keys for AI providers and cloud accounts, so a short window can expose a lot of secrets. Anyone who installed 1.107.0 or pulled the latest Docker images that day should treat the machine as compromised, remove the malicious files, go back to 1.106.2 and rotate every credential the machine could reach.
GitLab has fixed CVE-2026-110264, rated 10.0 out of 10, which under certain conditions could let an attacker with no account run code on a self-managed GitLab server. The bug is an integer overflow in the regular-expression engine GitLab ships, and it affects all versions from 7.0 onwards. Fixed versions are 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 and 18.11.13; GitLab.com is already patched.
Why it matters: GitLab servers often hold an organisation's source code and deployment secrets. Servers patched on 12 October for a similar bug are still vulnerable to this one, and GitLab has offered no workaround, so self-managed admins should upgrade now. There is no sign of attacks yet.
Hackers broke into the infrastructure of Onamae Desktop Cloud, a Japanese virtual server service run by GMO Internet, and installed malicious programs on up to 26,739 customers' servers. The company began shutting down customer servers from 4am local time on 21 October, and all Desktop Cloud plans are affected with no restart date yet. Data stored on the servers may have been exposed, though payment card details are kept elsewhere and were not affected.
Why it matters: When a hosting provider itself is breached, every customer is exposed at once, however careful they were. Customers should change any passwords or keys stored on their servers and watch for phishing emails that use the incident as a lure.
India's largest bank told stock exchanges that an early review of about 375,000 records posted on the dark web found 3,897 containing customers' personal details, including names, mobile numbers, dates of birth, email and home addresses and loan-related data. The bank was alerted on 16 October by outside agencies and has brought in an independent investigator to work out where the data came from. It says its operations are not affected.
Why it matters: Leaked details like these are exactly what scammers use to pose convincingly as bank staff. Digital rights group the Internet Freedom Foundation has urged the bank to tell affected customers directly, noting Indian law does not yet require it to.
Rotterdam police say they have arrested two men suspected of running a new kind of bank helpdesk fraud that cost about 170 victims around €380,000 in three weeks. Victims got AI-generated calls warning of suspicious account activity, were put through to a fake bank employee, and were told to install an app and hold their bank card against their phone. The app relayed the card's contactless chip data to an accomplice waiting at a cash machine.
Why it matters: This is a new twist on phone scams that turns a victim's own phone into a card skimmer. Police stress that your bank will never ask you to download an app, hold your card to your phone or share your PIN: if a call feels wrong, hang up and ring your bank on its official number.
Kaspersky says North Korea's Lazarus Group is using ML-KEM, a new post-quantum encryption standard, to protect the delivery of malware in a campaign it calls Operation FlashHole. The attacks, running since July 2025, target South Korean organisations through phishing pages and a popular local security tool, and run mostly in memory to avoid detection. At least three companies in software, construction and telecoms equipment were hit.
Why it matters: State hackers are quickly adopting cutting-edge cryptography to make their traffic harder to inspect. For defenders it is a reminder that spotting attacks increasingly depends on watching behaviour on devices, not just reading network traffic.
