AI is discovering new software vulnerabilities at a rapid pace. That sounds like good news, but attackers can then also easily exploit those vulnerabilities.
AI is causing an explosion of newly discovered software vulnerabilities. Where security researchers previously needed weeks or months to find complex bugs, specialized AI models are now taking over a large part of that investigative work. This leads to a striking trend: the number of registered vulnerabilities is rising faster than ever, while software vendors are simultaneously releasing a record number of security updates.
According to figures from the National Vulnerability Database (NVD), 45,207 new vulnerabilities were already registered between January and the end of July. If this pace continues, 2026 will see approximately twice as many security flaws as last year. At first glance, that seems like bad news, but security experts primarily see a positive evolution: AI helps to expose existing weak spots faster, even before cybercriminals can get away with them.
AI accelerates the search for vulnerabilities
The impact is also visible among software vendors. During its July Critical Patch Update, Oracle repaired 1,449 vulnerabilities, compared to 309 a year earlier. Microsoft, Google, and other major players are reporting a sharp increase in the number of security patches.
According to researchers, this does not mean that software has suddenly become worse. The primary explanation is that AI drastically increases the efficiency of security research. Models can analyze millions of lines of source code, recognize patterns, and identify potential vulnerabilities in a fraction of the time human researchers would need.
AI is increasingly evolving from a developer assistant into a specialized partner that actively assists with vulnerability research and secure coding.
Cybersecurity becomes an AI arms race
Various AI companies are now developing models specifically aimed at cybersecurity. Anthropic set the tone earlier this year with Mythos, a model that the company says performs exceptionally well in detecting unknown vulnerabilities. OpenAI, Google, and Microsoft are investing heavily in AI for security research. Microsoft is working on Project Perception, among others, while Mozilla previously indicated it had found and resolved significantly more security flaws thanks to Anthropic Mythos.

However, the race is not just about defense. The better AI becomes at finding vulnerabilities, the more interesting the technology becomes for attackers.
Cybercriminals also gain more powerful tools
Security experts have long warned that AI accelerates both defensive and offensive cyber capabilities. Advanced models can analyze vulnerabilities, build exploit chains, and automate attack scenarios. Intelligence services within the Five Eyes alliance therefore expect AI to fundamentally change the nature of cyber threats in the coming period.
Those warnings received extra attention recently after OpenAI announced that an experimental AI model independently carried out an attack on Hugging Face during internal safety tests after managing to escape its sandbox. Although it was a controlled test, the incident underscores how powerful autonomous AI systems are becoming and why AI security is becoming increasingly important.
More vulnerabilities also means more work
For IT departments, the rising stream of vulnerabilities has a clear downside. More discovered security flaws also mean more patches, more prioritization, and greater pressure on vulnerability management. Companies will therefore have to further automate their patch management and risk assessment. After all, not every vulnerability requires immediate action, but it is becoming increasingly difficult to manually determine which updates are a priority.
