AI hackers outperforming humans at a fraction of the cost
What's the story
A recent study by Stanford University has revealed that artificial intelligence (AI) can outperform human hackers in certain scenarios. The research focused on an AI agent called ARTEMIS, developed by Stanford researchers to test computer security. The team gave ARTEMIS access to Stanford's computer science network, which consists of around 8,000 devices including servers, computers, and smart systems.
AI advantage
ARTEMIS's performance surpasses human cybersecurity experts
The researchers allowed ARTEMIS to work for 16 hours over two days, while human cybersecurity experts were also given at least 10 hours to look for issues. The results showed that within this time, ARTEMIS identified nine real security vulnerabilities with a high degree of accuracy. It outperformed nine out of 10 professional hackers in the test and came second overall.
AI efficiency
Unique approach and cost-effectiveness
When ARTEMIS detects something unusual, it can create additional background tasks to probe multiple targets simultaneously. This multitasking capability gives the AI an edge over human testers who have to check things one by one. Plus, running ARTEMIS costs only $18 per hour, much cheaper than hiring a professional penetration tester who can earn around $125,000 annually. Even a more advanced version costs $59 per hour.
AI drawbacks
Limitations and the evolving hacking landscape
ARTEMIS struggles with tasks that require clicking through visual screens and sometimes overlooks serious issues. It also tends to raise false alarms, mistaking harmless activity for a successful attack. The study comes as hackers are already using AI tools for phishing emails, fake identities, and even gaining access to company systems.