AI Agents Find Security Flaws with High Precision
New scientific research documents that artificial intelligence can identify 77% of all security vulnerabilities in real-world software. In a comprehensive test, a specially developed AI agent systematically scanned and found critical flaws that hackers can exploit to break into systems.
The research confirms what security experts have long feared: The same AI technology used to protect systems can also be used by criminals to attack them. And according to the study, this is already happening today.
Cybercriminals Actively Using the Technology
Criminal groups and state-linked attackers have already adopted general-purpose AI for their operations, according to researchers. The technology is primarily used for two purposes: scanning for vulnerabilities in targeted systems and developing malware.
This represents a paradigm shift in cybercrime. Previously, finding security flaws required deep technical expertise and many hours of manual work. Now AI agents can automate the process and work around the clock with high precision.
From Theoretical Threat to Operational Reality
A success rate of 77% means the technology is far beyond the theoretical stage. For comparison, even experienced security researchers rarely find all vulnerabilities in complex systems.
"It's no longer a question of whether AI can be used for hacking, but how widespread the use already is," state the researchers behind the February 2026 study.
The ability to systematically scan large amounts of source code and identify weaknesses makes AI agents an extremely effective tool for attackers. It also lowers the threshold for less sophisticated criminal groups to carry out advanced attacks.
Consequences for Cybersecurity
For businesses and authorities, this development means the threat landscape has fundamentally changed. Traditional security measures based on attackers having limited resources and time no longer hold.
AI-driven attacks can scale almost without limitation and be executed faster than human security teams can respond. This forces organizations to invest in their own AI-based cybersecurity solutions to keep pace.
State-Sponsored Actors Leading the Way
The study indicates that state-linked hacker groups in particular are early adopters of AI agents in their operations. These actors have both the technical competencies and financial resources to develop sophisticated AI tools.
For smaller criminal groups and individuals, the availability of AI-based hacking tools is expected to increase significantly in coming years as the technology becomes more widespread and easier to use.
The Race for AI Security
The research underscores that a race is now underway between defenders and attackers to exploit AI's potential. While security firms work to use AI to find and fix vulnerabilities faster, criminals use the same technology to find and exploit them.
This dual-use potential makes AI one of the most pressing challenges in cybersecurity in the coming years.
