Sunday, November 22, 2026 · 1 MIN READ
AI Data Exposure and Exploited Flaws Put Business Trust to the Test
IT BRIEFING FOR BUSINESSES
Today’s reports put two business IT priorities in focus: limiting what AI services can access and closing vulnerabilities in the systems behind everyday work. Researchers reported stolen prompts at AI hosting providers, including requests marked for zero retention, while a laboratory attack against Meta Muse exposed private conversations and stored memories. Review which AI tools staff use, what business information they submit, and which accounts those tools can reach. For Microsoft 365, email, and other connected services, keep permissions narrowly scoped rather than assuming an assistant’s privacy settings prevent every route to exposure. Signal’s planned screenshot restrictions also underline the need to review AI access on staff devices, not just in browsers.
On infrastructure, prioritize checking for the actively exploited flaws affecting VMware vCenter, Cisco Expressway, SAP NetWeaver, and Apache Roller. Ask IT providers to verify whether document-processing applications depend on Ghostscript and whether encryption integrations use vulnerable GPGME versions; both projects have released fixes. Include network-facing systems and less-visible software components in that review. The ransomware-recovery fraud charge is a separate reminder to scrutinize recovery providers’ claims: require clear disclosure of their methods and charges, and test backups so the business understands its restoration options before an incident.
AI Providers and Assistants Face Data-Exposure Findings
OpenRouter dropped three AI inference providers after researchers reported months of undetected data theft, including requests marked Zero Data Retention. Separately, researchers demonstrated that a malicious webpage could extract private conversations and stored memories from Meta Muse without an approval prompt, with potential access to connected-account credentials. The Muse demonstration occurred in a laboratory and did not target real users.
SOURCES:The Hacker News The Hacker News
Ghostscript and GPGME Receive Security Fixes
Researchers demonstrated a patched Ghostscript flaw that allowed crafted uploads to escape its sandbox and execute commands on the processing server. GnuPG also released fixes for CVE-2026-129975, a critical memory flaw affecting GPGME versions before 2.3.0 that could expose application data when processing an OpenPGP key with an unusually long user ID.
SOURCES:The Hacker News The Hacker News
Attackers Use AI Agents Against Medical-Imaging Server
Amazon Threat Intelligence reported that suspected state-linked hackers used autonomous AI agents to discover an unknown MONAI Label vulnerability and obtain root access to a university server. The flaw permits unauthenticated file writes outside the server’s storage folder and affects version 0.8.5 and earlier in the default configuration.
SOURCES:The Hacker News
Signal Announces Screenshot Restrictions Against AI Access
Signal announced default screenshot restrictions intended to prevent AI agents from reading private conversations on mobile devices. The feature covers agentic systems in iOS 27 and Android versions where Signal can detect an AI assistant attempting to read chats; the company acknowledged that the protection is incomplete.
SOURCES:The Hacker News
Four Enterprise Software Flaws Added to Exploitation Catalog
CISA added four vulnerabilities affecting Apache Roller, Cisco Expressway, SAP NetWeaver, and Broadcom VMware vCenter to its Known Exploited Vulnerabilities catalog after finding evidence of active exploitation.
SOURCES:The Hacker News
Ransomware Recovery Executive Charged With Fraud
The chief technology officer and co-founder of ransomwarehelp.com was charged with wire fraud following an FBI sting. The company claimed to restore encrypted data using advanced cryptographic capabilities, but investigators alleged that it instead paid attackers for decryption keys and charged victims a markup for work it had not performed.
SOURCES:Krebs on Security