AI Threat AlertSocial Engineering
Generative AI lowers the cost of social engineering by orders of magnitude. Spear-phishing emails that previously required a fluent writer and target research are now produced in seconds with reasonable per-target personalisation. Voice cloning (ElevenLabs, OpenVoice, and others) enables real-time impersonation of executives and family members; multiple confirmed business-email-compromise and CFO-fraud incidents in 2023-2024 used cloned voices. Deepfake video is good enough for short verification clips and live calls under poor video conditions. Beyond direct attacks, AI-generated content fuels disinformation campaigns, fake review economies, and pig-butchering scams at unprecedented scale. AI Threat Alert tracks this category through CVEs in voice/face-recognition systems that fail to detect synthetic media, plus incidents in AIID (the AI Incident Database). Defenses: out-of-band verification for sensitive actions, deepfake detection layered with provenance signals (C2PA), and user education that assumes any voice or video can be faked.
| Severity | CVE | Headline | Package | CVSS |
|---|---|---|---|---|
| HIGH | CVE-2026-86063 | Claude Code: unverified plugin SHA checkout enables code install | claude-code | - |
| LOW | CVE-2026-60008 | Gitea: notebook CSS class injection enables UI redress | 3.5 | |
| HIGH | CVE-2026-84312 | Nginx-UI: Casdoor SSO login CSRF forces victim into attacker session | 7.1 | |
| LOW | CVE-2026-76894 | ckan-mcp-server: missing postMessage origin checks in UI | @aborruso/ckan-mcp-server | 3.1 |
| MEDIUM | CVE-2026-76297 | Splunk Enterprise: CSRF via GET disables federated search | 5.4 | |
| MEDIUM | CVE-2026-79980 | Zephyr RTOS: predictable DHCPv4 xid enables gateway/DNS hijack | 5.6 | |
| MEDIUM | CVE-2026-118677 | n8n Local Gateway: silent deep-link re-pairing | n8n | 5.9 |
| HIGH | CVE-2026-107743 | Mapster WP Maps: unauth REST route enables stored XSS | 7.2 |
Page 7 of 7