ARTFEED — Contemporary Art Intelligence

arXiv Paper on Agentic AI Skills Corrects Errors, Reverses Key Finding

ai-technology · 2026-08-06

A newly updated paper on arXiv (2603.00195v2) focuses on the formal analysis and security of supply chains concerning agentic AI capabilities, rectifying errors in the bibliography and external assertions. The authors confirmed 22 references through the arXiv API and informed the relevant authors. Additionally, they amended three external statements: MalTool indicates 1,300 standalone and 5,727 embedded malicious tools (not 6,487); CVE-2026-25253 pertains to authentication-token exfiltration via an unvalidated gatewayUrl (attributed to depthfirst, resolved on 2026.1.29), rather than remote code execution; ClawHavoc reports counts of 341, 824, and 1,184 by source and date, not "over 1,200." All experiments were re-evaluated against the released v0.6.0 implementation using committed harnesses. While experiments E1/E2 remain consistent (F1 96.15%), E3 shows a negative result: information flow analysis does not enhance detection. The document comprises 32 pages, features 5 theorems with complete proofs, includes 68 references, and provides an open-source tool at https://github.com/qualixar/skillfortify.

Key facts

  • Paper arXiv:2603.00195v2 is a revised version correcting bibliography and external claims.
  • 22 bibliography entries had author lists that did not match the cited arXiv identifiers; all corrected and affected authors notified.
  • MalTool reports 1,300 standalone and 5,727 embedded malicious tools, not 6,487.
  • CVE-2026-25253 is authentication-token exfiltration via an unvalidated gatewayUrl, credited to depthfirst and fixed in 2026.1.29.
  • ClawHavoc counts are 341, later 824, and 1,184 by source and date, not "over 1,200".
  • All experiments re-measured against v0.6.0 implementation using harnesses committed to the repository.
  • E1/E2 unchanged (F1 96.15%).
  • E3 reverses to a negative result: information flow analysis adds no detection benefit.
  • Paper has 32 pages, 5 theorems with full proofs, 68 references.
  • Open-source tool available at https://github.com/qualixar/skillfortify.

Entities

Institutions

  • arXiv
  • GitHub
  • qualixar

Sources