US institutions: The number of cybersecurity vulnerabilities discovered by AI this year will be twice that of last year

On July 28, the number of software security vulnerabilities discovered in popular technology products in 2026 is expected to reach approximately double the number discovered in 2025. This surge is mainly due to the continuous improvement of the capabilities of artificial intelligence systems. The U.S. National Vulnerability Database shows that from January to Monday, a total of 45,207 vulnerabilities were recorded, which is close to the total for the whole of 2025. Last year, the database recorded a record number of vulnerabilities. Oracle (ORCL.N) said its monthly software updates in July fixed 1449 security vulnerabilities, a record high; updates for the same period last year included only 309 fixes. Microsoft (MSFT.O) disclosed 642 security vulnerabilities in July, also setting a record high, almost five times the number in the same period last year. Google (GOOG.O) discovered and fixed 433 such vulnerabilities in a recent Chrome update, compared with 11 in a similar update a year ago. "We must face the fact that these tools are enhancing people's ability to discover software vulnerabilities," said Gabriel Shapiro, a distinguished artificial intelligence research scientist at cybersecurity firm SentinelOne. Doug Turner, director of engineering at Google Chrome, said vulnerabilities were discovered at an "unprecedented scale and speed" thanks to advances in artificial intelligence models and corresponding investments.