AI Hacker: Stanford University’s computer network recently became a new battlefield in the ongoing invisible war between humans and Artificial Intelligence. This time the result was shocking because an AI agent left behind those professional cyber security experts who take salaries of lakhs for this work. This experiment clearly indicates that the world of cyber security is changing rapidly.
ARTEMIS: An AI that outwitted hackers
The AI agent ‘ARTEMIS’ used in this experiment was prepared by researchers from Stanford. It was left on the university’s private and public computer science networks for 16 hours. During this period, ARTEMIS scanned about 8,000 devices which included servers, computers and smart systems. By the end of the test, this AI proved to be better than nine out of ten professional penetration testers and caught many vulnerabilities that even experienced human hackers ignored.
List of research and results
This study was led by Stanford researchers Justin Lin, Elliot Jones and Donovan Jasper, who are experts in cybersecurity, AI agents and machine learning safety. According to him, ARTEMIS was designed to automatically scan and analyze complex systems without interruption for long periods of time, whereas common AI tools often fail in such multi-step tasks.
How was AI’s advantage over human hackers?
Ten experienced penetration testers were called for this experiment and all were asked to work for at least 10 hours. Comparison was made with the first 10 hours of ARTEMIS. During this period, AI discovered nine real security flaws with an accuracy of 82 percent. Researchers say that while other AI systems lag behind humans, ARTEMIS’s performance was on par with the strongest human participants.
Huge difference in expenses
The most surprising thing was its cost. According to the study, the cost of running ARTEMIS is around $18 per hour i.e. around Rs 1,600. Whereas in America, the annual salary of an average professional penetration tester is more than 1.25 lakh dollars, which is more than one crore in Indian Rupees. A more advanced version of ARTEMIS also does the job for about $59 per hour, which is still cheaper than hiring a top-level expert.
How Artemis discovered what humans missed
The biggest strength of ARTEMIS is its design. As soon as it finds something suspicious during a network scan, it immediately launches small ‘sub-agents’ which start investigating from different angles at the same time. In this way it is able to analyze multiple threats simultaneously which is almost impossible for humans.
In one case, this AI detected a major weakness in an old server that human testers had abandoned because their browser was unable to load it. ARTEMIS seamlessly accessed the system through a command-line interface.
Limitations of AI also came to light
However ARTEMIS is not completely perfect. It faltered in tasks that required working with a graphical user interface. A serious flaw was missed because the AI could not interact properly with simple clicks or visual elements. On some occasions, it even mistook useless network activity as a threat, meaning false positives also emerged.
According to the researchers, ARTEMIS is extremely fast and accurate in text- and code-based environments, but it still needs improvement where systems become more visual.
Also read:
Airtel Vs Jio: Which recharge will give the strongest speed and more benefits? Know who is superior to whom
