01 / AI · Analysis
Soongsil decoy servers block ARTEX AI tool in 60-server test
The test opponent was a tool disclosed as used in state-backed hacking, and public materials show no deployment on live production servers.

Researchers at Soongsil University, a private university in Seoul, said on 9 Nov 2026 that a decoy-server defence blocked AI hacking in a test on 60 web servers, a result published by the team itself. The team, led by Professor Choi Dae-seon, works at the university's AI Safety Research Center.
The opponent was an autonomous penetration AI disclosed as having been used in real state-backed hacking, according to a Korean press report. That choice gives the announcement its weight, because the defence was tested against a tool with disclosed real-world use.
The defence sends the attack tool to fake servers instead of real ones, the team said. The tool's record is the starting point for reading the figures.
A test opponent tied to state-backed hacking
The team chose ARTEX, an autonomous penetration AI. Microsoft disclosed last month that the tool had been used in state-backed hacking, according to the press report.
ARTEX was made public in July, the report said. It is a type of tool that finds vulnerabilities and breaks in without waiting for human instructions.
Sixty web servers, before and after the defence
Without the defence, ARTEX breached 35 of 60 vulnerable web servers, according to figures published by the team. With the defence applied, no attack reached a real server.
The team's baseline for comparison was the condition with no defence applied. Both results are team-published figures.
The paper containing the results was posted on arXiv, an open online repository of research papers, according to the report. The figures cover the 60-web-server experiment. Public materials include no case of the technology being applied to live production servers.