Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsResearch

Google: half of the vulnerabilities AI finds lead to remote code execution

Google Threat Intelligence Group found that 50 percent of the vulnerabilities AI tools discover enable remote code execution, against 26 percent of those found other ways. Monthly CVE disclosures doubled in 2026, from 5,045 in January to 10,740 in August, and 141 vulnerabilities were exploited in the wild from January to August, against 127 in all of 2025. Attackers used CVE-2026-1731, a BeyondTrust flaw an AI tool found, within four days of its disclosure. Google's own report, published September 30, also counts 2,076 AI-related CVEs from January 2025 to August 2026, half of them in agent orchestration middleware, where disclosures rose 347 percent in 2026. Google has not yet seen a zero-day used against AI infrastructure, but attackers are already exploiting newly disclosed flaws in exposed middleware; its examples include LiteLLM and Langflow.

Why it matters

AI now finds serious bugs faster than most teams patch them, and attackers read the same disclosures. Patch first what is being exploited, not whatever arrived first, and put internet-facing AI gateways and orchestration tools at the top of that list.

Sources

Read it with the rest of the news in the Learning Center