Skip to content
EventPast event

Soongsil University uses honeypots and shadow replicas to lure LLM attack agents

2 reports2 reporting sourcesLatest report

Start here

AI overview

The AI Security Research Center at Soongsil University proposes a deception-based defense scheme: an LLM reads the application source code and then pre-embeds fake pages, fake paths and fake passwords; visitors who touch the decoys are redirected to an isolated replica with an identical appearance, which the team calls the "Gingerbread House." The study uses the autonomous penetration AI ARTEX as the adversary; without defenses it compromised 35 of 60 vulnerable Web servers, and with the defense enabled the real servers were compromised 0 times; against an attack agent built on Claude Opus 4.8, the number of compromises dropped from 26 to 1. The decoys cost about 2 dollars per service and take 7 to 16 minutes for one-time deployment, and 117 normal crawler visits produced no false positives.

Generated by AI from reports · Updated 9 days ago

Attention and developments

Attention, hourly

Now 38, 0% in 24 hours·Peak 38 (Nov 13 04:00)

Nov 14Nov 15Nov 16Nov 17Nov 18Nov 19Nov 20Now 38

Progress timeline

Nov 9
  1. aisc.ssu.ac.kr
    Soongsil University AI Security Research Center develops deception defense technology that lures AI hackers

    The team at the AI Security Research Center of Soongsil University has developed a deception-based defense technology that lures autonomous-penetration AI hacker tools into fake copies isolated from real servers; the paper has been published on arXiv (2611.07611). The team used ARTEX, the autonomous penetration AI that Microsoft disclosed in October as being used in state-backed hacking attacks, as the opponent in experiments: without defenses it breached 35 of 60 vulnerable web servers, and with the defense enabled the real servers were breached 0 times, and against an attack Agent built with Claude Opus 4.8 it also dropped from 26 times to 1 time. The approach is to have the AI read its own service source code and then pre-embed fake pages, fake paths and fake passwords; visitors who touch the decoys are redirected to an isolated copy with an identical appearance, which the team calls Gingerbread House; the decoys cost about 2 USD per service and are deployed once in 7 to 16 minutes, with no false positives across 117 normal crawler visits.

Nov 5
  1. arXiv: Alignment, Deception & Oversight
    A gingerbread house for LLM agents: a honeypot planned from source code lures autonomous attackers into an isolated shadow copy

    Researchers propose a honeypot defense scheme against autonomous attacking agents, in which an LLM defender plans static honeypots based on the application source code, and a router forwards subsequent requests from clients confirmed as attackers, together with available exploits, to an isolated shadow copy. On 40 critical tasks in CVE-Bench, the static scheme reduced real-target compromises of Claude Opus 4.8 from 26/40 to 5/40, and reduced compromises by the ARTEX attack framework from 20/40 to 0/40 (31 valid runs); on 20 entirely new tasks that did not participate in the scheme design, ARTEX went from 15/20 to 0/20. In about half of the deception-group runs, ARTEX's exploits succeeded on the shadow copy while the real target remained intact, indicating that the defense diverts rather than prevents attacks, and the residual risk lies in requests that reach the real target before confirmation is complete.

Related discussion

Total 1 items

Attention trend

New sources per day

2 in 11 days·most 1 (Nov 10)·today 0

  • Nov 10: 1 new sources (0 outlets, 1 accounts)
  • Nov 11: 1 new sources (1 outlets, 0 accounts)
  • Nov 12: 0 new sources (0 outlets, 0 accounts)
  • Nov 13: 0 new sources (0 outlets, 0 accounts)
  • Nov 14: 0 new sources (0 outlets, 0 accounts)
  • Nov 15: 0 new sources (0 outlets, 0 accounts)
  • Nov 16: 0 new sources (0 outlets, 0 accounts)
  • Nov 17: 0 new sources (0 outlets, 0 accounts)
  • Nov 18: 0 new sources (0 outlets, 0 accounts)
  • Nov 19: 0 new sources (0 outlets, 0 accounts)
  • Nov 20: 0 new sources (0 outlets, 0 accounts)

Hourly attention

Current attention 38·Peak in the comparable period 38 (Nov 13 04:00)·Change over the comparable past 24 hours 0%

010203040Nov 1304:00Nov 1512:00Nov 1719:00Nov 2003:00