Storia in 2 fonti

Malicious MCP Servers Can Split Instructions to Make AI Coding Agents Exfiltrate Secrets

GhostSplice splits MCP instructions across channels, letting AI coding agents combine them to exfiltrate SSH keys, secrets, and source code.

Raccontata da

securityweek.com

thehackernews.com

Confronto fonti

2 prospettive sulla stessa storia

AI · summaries

thehackernews.comStai leggendo3 g fa

Malicious MCP Servers Can Split Instructions to Make AI Coding Agents Exfiltrate Secrets

GhostSplice splits MCP instructions across channels, letting AI coding agents combine them to exfiltrate SSH keys, secrets, and source code.

originale

securityweek.com3 g fa

'Ghostjacking' Attack Uses Poisoned Logs to Turn AI Agents Bad

Tenet demonstrated Ghostjacking: malicious log injection on Cloudflare, Datadog, Sentry (50% Fortune 500 use them) hijacks AI agents 90% success rate—credentials stolen, DNS redirected. For tech leads: identifies pattern where AI agents blindly trust logs/alerts—requires security review of observability tool configs and autonomous AI governance before enterprise rollout.

Leggi questa versione →

Timeline cronologica

lunedì 10 agosto 2026·securityweek.com
'Ghostjacking' Attack Uses Poisoned Logs to Turn AI Agents Bad
In a Ghostjacking attack, an AI agent executes instructions planted in the log that records a blocked request word for word.
martedì 11 agosto 2026·thehackernews.com
Malicious MCP Servers Can Split Instructions to Make AI Coding Agents Exfiltrate Secrets
GhostSplice splits MCP instructions across channels, letting AI coding agents combine them to exfiltrate SSH keys, secrets, and source code.