GOOGLE Threat Intelligence Group (GTIG) says vulnerability disclosures and exploitation both accelerated during 2026, although actively exploited flaws remain a small proportion of the total. Disclosures rose from 5,045 in January to 10,477 in July and 10,740 in August. GTIG recorded 141 distinct vulnerabilities disclosed and exploited between January and August, compared with 127 during all of 2025, while the monthly average increased from 10.5 to 18.
Only 0.23% of vulnerabilities disclosed in 2026 were observed being exploited. Zero-day exploitation rose more modestly, from an average of eight per month in 2025 to 11 in 2026, reaching 22 in August. GTIG attributes most of the wider increase to the rapid weaponisation of previously disclosed, or “n-day”, flaws. High-Risk vulnerabilities exploited rose from 28 in 2025 to 75 between January and August 2026.
The research also identifies a different risk profile for vulnerabilities found with AI assistance. Among the flaws GTIG could identify, 58% were rated Moderate risk and 4% High risk, compared with 28% and 3% respectively for vulnerabilities not attributed to AI. Half of AI-discovered flaws resulted in remote code execution (RCE), versus 26% across the broader CVE dataset, although GTIG cautions that public attribution data is incomplete.
The report cites CVE-2026-1731, an unauthenticated command-injection flaw in BeyondTrust Privileged Remote Access and Remote Support, which GTIG says was exploited within four days of disclosure.
GTIG tracked 2,076 AI-related CVE disclosures from January 2025 to August 2026, including more than 1,500 in 2026. AI orchestration frameworks accounted for 782 2026 vulnerabilities, followed by inference and serving infrastructure with 212. Confirmed exploited examples include CVE-2026-42271 in LiteLLM, CVE-2026-5027 in Langflow and CVE-2025-3248 in Langflow. The group recommends threat-intelligence-led prioritisation, targeted protection of exposed services and faster, automated remediation.