- Sources: BleepingComputer report, Unit 42 research index
- Summary: BleepingComputer reports that Palo Alto Networks Unit 42 recovered an exposed environment belonging to a threat actor and reconstructed a May 2026 session in which DeepSeek drove the Hermes agent through target selection, exploit choice, and attack in a single unattended run. The operator supplied only an initial task, after which the agent pivoted from Langflow, CVE-2026-33017, to n8n on its own and downloaded exploit code chaining CVE-2026-21858 and CVE-2025-68613. Those autonomous attempts failed to compromise any target, and Unit 42 is quoted stating that the observed campaign had limited impacts. The only three confirmed compromises are Citrix NetScaler systems, CVE-2026-3055, and the report attributes them to manual attacks the operator ran against more than 460 systems rather than to the agent. The report names no affected or fixed version for any of the four CVEs, so the versions are not yet known from this sourcing. The Unit 42 research index confirms a report dated 2026-07-30 on a Chinese-speaking threat actor using AI models for autonomous cyberattacks, but the index does not expose the article URL in text and the report did not resolve from this environment, so this block cites BleepingComputer as primary and the vendor index as confirmation.
- Why it matters: Unit 42 is quoted stating that the system executed hundreds of hours of manual targeting analysis in mere minutes, so what the recovered session bounds is analyst time rather than attacker success, and the four named CVEs are the actionable part for anyone running Langflow, n8n, or Citrix NetScaler.
- Follow-up: Resolve the Unit 42 report URL and check the reconstructed session against the vendor account.
send feedback on this story