Qualys Says AI Is Accelerating Exploitation of Familiar Security Weaknesses

Summary
Mandiant's 2026 report says attackers are using agentic AI, but exploitation of existing weaknesses remains a leading intrusion route. Qualys argues teams should inventory AI assets, validate exposures, shorten remediation time and add runtime monitoring.
Key points
- Mandiant and Google Threat Intelligence Group describe attackers using agentic AI for reconnaissance, vulnerability validation and credential harvesting; GTIG assessed with high confidence that an AI model helped find and weaponize a two-factor authentication bypass.
- M-Trends 2026 attributes 32% of intrusions to exploits and estimates mean time to exploit at minus seven days.
- In six of the report’s eight case studies, attackers used exposures that could have been identified beforehand, including poisoned packages, long-lived CI/CD credentials, unsigned plugins and over-privileged agents.
- Qualys says its analysis of the Hugging Face Kubernetes intrusion counted about 17,600 agent actions over five days; the agent reached cluster-admin in about 13 hours after entering through a dataset configuration file and unsanitized template.
- The article cites AI-component vulnerabilities CVE-2025-23266 and CVE-2025-6514, for which Qualys research puts weaponization before disclosure, and CVE-2026-28400, whose exploitability depends on runtime conditions.
- Recommended steps include inventorying AI assets, replacing agents’ long-lived credentials, testing models and MCP servers, validating exploitability, and monitoring AI workloads at runtime.
- Qualys presents its own products as supporting AI asset discovery, exploit validation, remediation and runtime detection; these are vendor-described capabilities.
Article Details
- Topic
- AI-enabled cyber threats, AI infrastructure exposures, and risk operations for detecting and remediating vulnerabilities
MITRE ATT&CK
CVE
CVE-2025-23266NVIDIA · Container ToolkitQualys research on AI exploits in containers puts the window of weaponization at minus one day for flaws in the NVIDIA Container Toolkit (CVE-2025-23266) and in an MCP server (CVE-2025-6514).CVE-2025-6514Vendor / Product unavailableQualys research on AI exploits in containers puts the window of weaponization at minus one day for flaws in the NVIDIA Container Toolkit (CVE-2025-23266) and in an MCP server (CVE-2025-6514).CVE-2026-124194Microsoft · Microsoft 365 Apps for Enterprise; Microsoft · Microsoft Office LTSC 2021; +1 moreMicrosoft fixed more than 1,000 vulnerabilities, 63 of them critical, including an exploited Outlook spoofing zero-day (CVE-2026-124194) with a CISA patch deadline of November 24.CVE-2026-28400Vendor / Product unavailableFor a Docker Model Runner flaw (CVE-2026-28400), whether a workload is exploitable depends on runtime flags and how the service is invoked.CVE-2026-96951Microsoft · Microsoft Edge (Chromium-based); Microsoft · Outlook.com; +18 moreMicrosoft fixed a record 1,414 vulnerabilities, 78 of them critical, including an exploited Windows Shell zero-day (CVE-2026-96951) with a CISA patch deadline of October 16.
People
Malware
Vendors
MandiantMandiant’s AI Risk and Resilience 2026 report shows attackers putting agentic AI to work across the intrusion.MicrosoftMicrosoft fixed a record 1,414 vulnerabilities, 78 of them critical, including an exploited Windows Shell zero-day (CVE-2026-96951) with a CISA patch deadline of October 16.NvidiaQualys research on AI exploits in containers puts the window of weaponization at minus one day for flaws in the NVIDIA Container Toolkit (CVE-2025-23266) and in an MCP server (CVE-2025-6514).QualysQualys AI agents detect new exposures, prove which ones are exploitable, and remediate them, so the defender’s side of the race is automated too.
Products
Container Runtime SecurityDetect AI Exploitation in Real Time with Container Runtime SecurityDocker Model RunnerFor a Docker Model Runner flaw (CVE-2026-28400), whether a workload is exploitable depends on runtime flags and how the service is invoked.ETM IdentityTreat Every Agent as an Identity with ETM IdentityNvidia Container ToolkitQualys research on AI exploits in containers puts the window of weaponization at minus one day for flaws in the NVIDIA Container Toolkit (CVE-2025-23266) and in an MCP server (CVE-2025-6514).Qualys Enterprise TruRisk Management (ETM)InstaScan, powered by Agent Insta in Qualys Enterprise TruRisk Management (ETM), matches new advisories against the inventory Qualys already holds, without a rescan.Qualys InstaScanInstaScan, powered by Agent Insta in Qualys Enterprise TruRisk Management (ETM), matches new advisories against the inventory Qualys already holds, without a rescan.TotalAISee AI as Attack Surface with TotalAITotalCloudunknown malware in container images and workloads before deployment and in cloud network traffic with TotalCloud Cloud Detection and Response.TruConfirmAgent Val uses TruConfirm to safely prove whether an exposure is exploitable in your environment, drives the mitigation, and revalidates closure.TruRisk EliminateAgent Sara plans and runs remediation in TruRisk Eliminate.
