[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Context & Harness Attacks · markdown

research (NebuSec write-ups, Nov 20, 2026); fixes per the researchers, Meta has not commented publicly · published here

NebuSec chained a malicious web page in Meta Muse's PDF export into silent theft of chats and memories, then a kernel escape that reached the user VM's credential store; researchers say all of it is patched

  • For agent builders: egress approval that trusts 'reads' can be drained through GETs; treat query strings and paths as data. Keep browser sandboxes on even in helper tasks like PDF export, and don't let the agent's container share a kernel with your secrets.

Sources: NebuSec: Hacking Muse: Stealing Private Chats Through Browser Use (Nov 20, 2026) · NebuSec: Inside Meta Muse: Browser, Sentinel and the Shared-Kernel Boundary (Nov 20, 2026)