{"uuid": "9eb8a13f-c526-4bd8-9333-7e10e00cb341", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-97228", "type": "seen", "source": "https://bsky.app/profile/nexttool.bsky.social/post/3mx6nnoagry2w", "content": "3/6 A structural MCP flaw: prompt-injection can hop from one internal AI agent to another via implicit agent-to-agent trust, enabling SSRF and data exfil. Bugs hit Google (sev 8.0) and Rapid7 (CVE-2026-97228, fixed). Treat LLM\u2192tool calls like untrusted input. https://arstechnica.com/security/2026/10", "creation_timestamp": "2026-10-06T05:02:21.071316Z"}