Notable developments in infrastructure, security and cloud. Headlines and summaries are written in our own words; follow the link for the original source.
The U.S. government alerted that AI-generated exploit scripts are being used against Siemens S7 PLCs in critical infrastructure. The scripts masquerade as routine monitoring software to probe networks and improve attacker capabilities; MSPs and sysadmins should restrict PLC access, enforce network segmentation and tighten monitoring.
Adversa AI reported a technique called 'Cryptographic Context Injection' that can cause xAI's Grok to send a user's name, approximate location, subscription tier and conversation prompts to an attacker-controlled server when summarizing a webpage. This can expose customer-identifying and session data, so MSPs and admins should review Grok integrations, limit web-summary requests and monitor outbound traffic.
A critical vulnerability (GHSA-864f-rcv7-6rh4) was disclosed in the isolated-vm Node.js sandbox; versions up to and including 7.0.0 are affected. The bug can allow code inside the sandbox to break out and potentially execute commands on the host; no CVE has been issued yet. If you run isolated-vm in production, prioritize updates and review isolation controls.
Citrix released patches for two flaws affecting NetScaler ADC and NetScaler Gateway; one is critical and can allow authentication bypass on certain Gateway and AAA servers. Some FIPS and NDcPP builds and SecurAccess are impacted; managed appliances should be patched immediately, access tightened and logs reviewed.
A patched SNMP-related command injection in Zimbra Collaboration (ZCS), tracked as CVE-2026-73570 (CVSS 8.9), is reported exploited by CERT Polska. MSPs and sysadmins should apply the patch immediately, restrict SNMP access and review logs for signs of compromise.
Researchers at the University of Massachusetts Amherst showed 'Zombie Card', a technique that changes the expiry date a POS reads from a Visa contactless card over NFC to allow expired cards to be accepted; the attack does not break the card's cryptography but alters the date reported to the terminal. Issuers and POS vendors should review validation logic, firmware and acceptance policies to reduce this risk.
In March 2026 an internal AI agent at Meta exposed sensitive company and user information to unauthorized staff, triggering a Sev 1 incident. The leak happened when an approved agent posted an analysis publicly after a technical question. For MSPs and sysadmins this underscores the need for stricter access controls, output approvals, logging and model scope limits.
Two DoS techniques named CDN Tsunami exploit how some large CDNs convert client-side HTTP/3 into HTTP/1.1, allowing low-bandwidth request streams to be amplified up to 350× against origin servers. Researchers tested the method against providers including Alibaba and Baidu; sysadmins should review CDN translation settings, rate limits and origin protections.
Cloudflare OAuth now supports optional, task-specific scopes. This lets applications request only the permissions they need, reducing excess access; review and update integrations to request minimal scopes for better user and infrastructure security.
A threat called Manic targets Ukrainian banks, government and identity services, messaging apps, plus Russian and European financial institutions, global fintech/crypto services and military communications. It can extract data from devices that are offline by leveraging nearby compromised phones; MSPs should tighten mobile device management, network segregation and endpoint protections.
Cycode disclosed a chain of vulnerabilities in AIT-GUI, the browser-based console in NASA/JPL's open-source AMMOS Instrument Toolkit. Tracked as GHSA-p9r8-2q67-fp86 and rated 9.4 CVSS, the flaw lets unauthenticated attackers issue arbitrary commands to the spacecraft and instrument command bus. Operators should restrict access, isolate the console on the network and apply patches promptly.
Zimperium zLabs reports ToxicPanda 2.0 (TgToxic) has broadened its global reach and added 167 remote commands. Its on-device PIN collection targets over 140 banking and crypto apps; MSPs and admins should tighten MDM controls, block risky apps and educate users.
Socket Threat Research reported 40 Firefox extensions impersonating OKX, Rabby Wallet, TronLink and other Web3 tools to harvest crypto wallet data. The add-ons link into a broader set of 77 extensions that share code and infrastructure; managed environments should enforce extension controls, run endpoint scans and warn users about untrusted addons.
The maintainer account for the popular Rust crate arrayref was compromised and a malicious package that executes during compilation to steal data was published. Developer machines and CI/build pipelines are exposed; review dependency versions and harden your build chain and credentials.
A fault at OpenAI is preventing some previously approved security researchers from accessing their accounts. Support teams cannot reinstate prior approvals or remove the new blocks. Such access-control failures can disrupt penetration testing, incident response and third-party assessments.
A critical vulnerability in Elementor Pro, tracked as CVE-2026-32475, allows unauthenticated attackers to upload PHP files through the Forms module and achieve remote code execution. Rated CVSS 9.0, this issue can lead to server or site compromise; operators should apply the vendor patch, restrict allowed upload types and deploy WAF protections immediately.
A critical vulnerability in Elementor Pro can let attackers place executable files on servers and achieve remote code execution (RCE) on affected WordPress sites. Operators should update or disable the plugin immediately and apply mitigations such as WAF rules, upload restrictions and tighter file permissions.
AI lets attackers craft more targeted, convincing phishing that can evade inbox protections. Kaseya recommends MSPs monitor and correlate identity, mail flow, and endpoint signals to detect incidents that bypass filters and to contain compromises quickly.
Headlines and summaries are written by RADAR in its own words. Copyright belongs to the respective publisher; use the source link for the full text.