The Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability Exploitation
VELA benchmark
Researchers investigate how AI agents delegate tasks and interact with web backend interfaces, reshaping web vulnerability exploitation. The study evaluates commercial agents across a benchmark of 44 vulnerabilities.
why this severityDemonstrated by researchers with no known real-world exploitation.
- first report
- 27 Oct 2026
- last update
- 27 Oct 2026
Timeline · 1
Sources · 1
Research and catalogues
- 2026-10-27The Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability ExploitationarXiv · arxiv.org/abs/2610.35541
Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.
Classification
- lane
- Research
- AI role
- AI-enabled attack
- categories
- AI-assisted intrusion
- severity
- 1 · Research / proof of concept
- targets
- Framework / tooling
- owasp llm top 10
- no entry for these categories
- mitre atlas
- no technique for these categories
- dates
- happened 27 Oct 2026 · first reported 27 Oct 2026
- origin
- News & research feeds
- record
- confidence 0.90 · created 2026-10-27 15:31Z · updated 2026-11-11 00:12Z
Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned
Copy a brief
This record as a short Markdown brief, for a ticket, a chat or a coding agent.
the brief as text
# The Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability Exploitation https://gcdradar.vercel.app/incidents/the-agent-in-the-middle-how-ai-mediated-access-reshapes-web · AIR-2026-1569 · closed 2026-11-10 - Lane: Research - Severity: 1 (Research / proof of concept): Demonstrated by researchers with no known real-world exploitation. - Fix status: Fix status unknown - First reported: 2026-10-27 - AI role: AI-enabled attack - Products: VELA benchmark ## Summary Researchers investigate how AI agents delegate tasks and interact with web backend interfaces, reshaping web vulnerability exploitation. The study evaluates commercial agents across a benchmark of 44 vulnerabilities. ## What to do No action recorded; read the sources below and the vendor's advisory. ## Latest updates - 2026-10-27: The Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability Exploitation (https://arxiv.org/abs/2610.35541) ## Sources - 2026-10-27 arXiv: The Agent in the Middle: How AI-Mediated Access Reshapes Web Vulnerability Exploitation https://arxiv.org/abs/2610.35541 Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.
Related
- closedCyberDual CyberDual: AI Attackers versus AI Code Reviewers in Real Software Maintenance
Research · since 27 Oct 2026
AI - closedClaude Mythos 5 AISI Evaluation Agents Conduct Unsanctioned Real-World Cyber Activity
Research · since 25 Jul 2026
AI - closedGPT-5.6 Sol Autonomous OpenAI Evaluation Agents Compromise Hugging Face Infrastructure
Breaches & attacks · Exploited · since 8 Jul 2026
AI - closedDeepSeek Threat Actor Uses DeepSeek-Powered Hermes Agent for Exploitation
Breaches & attacks · Exploited · since 7 May 2026
AI - closedHermes Multi-Agent Framework Compromises Taiwanese Government Systems
Breaches & attacks · Exploited · since 1 Jul 2026
AI - closedOpenAI models OpenAI Models Escaped Sandbox and Reached Hugging Face Infrastructure
Research · since 11 Jul 2026
AI