The source-planned honeypot study tests containment rather than whether an attacker can exploit anything at all. A defender prepares a credible shadow application from source code, then a router moves identified hostile sessions away from the real service. Exploitation can succeed inside the shadow without changing the protected application's state. In the…
Inspect the ClaimsCase · DiggingBeagle record
ARTEX honeypot study separates successful decoy exploits from real compromise
A source-planned router redirected hostile sessions into an isolated shadow application. No primary compromise was recorded in the reported ARTEX deception arms, but the benchmark does not establish zero deployment risk.
Public disclosure: Nov 5, 2026
Scope
Research reported in a controlled research or disclosure setting. Exact prerequisites, versions and absence of documented exploitation must be preserved.
Assessment profile
Assessment method- Setting
- evaluation
- Exploitation
- demonstrated
- Evidence
- Grade C
- Remediation
- not applicable
- AI role
- AGAINST AI
Basis and provenance
The cited first-party advisory or research establishes the stated mechanism only within its documented conditions. No unrelated production compromise, extended victim scope, financial harm or universal mitigation is inferred.
Assessed Nov 9, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-ARTEX-DECOY-REDIRECT · Source-planned deception against ARTEX exploitation agents
At a glance
A relevant outcome is whether the protected service was compromised, not whether the attacker received a convincing success response.
Read the implicationsSelected single-service tasks and one retained run per condition; no field protection rate, independent high-volume benign baseline or adaptive client-identity evaluation.
Limits and uncertaintyFull account
The source-planned honeypot study tests containment rather than whether an attacker can exploit anything at all. A defender prepares a credible shadow application from source code, then a router moves identified hostile sessions away from the real service. Exploitation can succeed inside the shadow without changing the protected application's state.
In the ARTEX controls, the primary service was compromised on 20 of 40 CVE-Bench tasks and 15 of 20 alternative tasks. The corresponding static-deception arms recorded no primary compromise. The CVE-Bench static arm had 31 valid runs, however; the nominal 40-task denominator must not be mistaken for 40 valid independent demonstrations. Some shadow exploits still succeeded.
Where the protection resides
The router must identify a hostile client before consequential primary-side effects, keep shadow and primary state separate, and avoid diverting legitimate users. Some attack traffic reached the primary before the session was classified. The paper's benign crawler did not establish an independent, high-volume false-positive rate.
These are selected benchmark outcomes with one retained run per condition, validity exclusions and statistical uncertainty. They do not prove field effectiveness against adaptive identity changes. The relevant operational check is the protected service's independently measured state, not a convincing success response returned by the decoy.
Mechanism and trust boundary
- 01
Defender first inspects the application source and prepares a credible decoy representation.
- 02
The front router detects an exploitation attempt and routes later requests from that session to an isolated shadow service.
- 03
Subsequent exploit attempts may succeed inside the decoy while not reaching the protected system.
Claims & evidence
CLM-ARTEX-DECOY-REDIRECTThe study reports that a source-informed router moved detected ARTEX exploit traffic to a shadow service in the evaluated tasks.supported
Basis: reported finding
- supportsSource-planned deception against ARTEX exploitation agentsprimary disclosure
Abstract, Threat model and Results tables describing ARTEX attacker success
CLM-ENR231-ARTEX-REAL-VS-SHADOWIn the paper's ARTEX experiment, the unprotected control compromised the real target on 20 of 40 CVE-Bench tasks and 15 of 20 alt-cve-bench tasks; the evaluated deception arms recorded no real-target compromise, while some shadow exploits succeeded.supported
Basis: reported finding
- supportsSource-planned deception against ARTEX exploitation agentsprimary disclosure
Section 6.1 and 6.2, Table 3: real-target compromise and shadow-only success, including validity qualifications
Implications
A relevant outcome is whether the protected service was compromised, not whether the attacker received a convincing success response.
Controls and mitigations
- Use authenticated test boundaries and measured false-positive costs; independently confirm attacker isolation instead of relying on the model-facing response.
Unknowns and contradictions
- Selected single-service tasks and one retained run per condition; no field protection rate, independent high-volume benign baseline or adaptive client-identity evaluation.
Sources and citation
Material revision history
- Nov 9, 2026 · Published version · first publication · revision 233