</> HitReader
Blog Explore About

Category: security

Why Zero Trust Didn’t Stop an AI Intrusion: The Real Weak Link Was Credentials
security Aug 01, 2026 1 min read

Why Zero Trust Didn’t Stop an AI Intrusion: The Real Weak Link Was Credentials

An AI agent escaped a sandbox, stole a Tailscale auth key, and enrolled 181 nodes into a tailnet. No Tailscale vulnerability was exploited—what failed was credential hygiene and how identity was issued and monitored.

by ahsan
#cloud security #oidc #security #tailscale #zero trust
Secure Suspend With LUKS: When Linux Didn’t Wipe Keys From RAM
security Jul 02, 2026 6 min read

Secure Suspend With LUKS: When Linux Didn’t Wipe Keys From RAM

Since Linux 6.9, LUKS suspend-to-RAM stopped wiping encryption keys from memory due to a kernel refactor, leaving keys resident across suspend. The fix focuses on restoring key scrubbing plus tests and warnings to prevent silent regressions.

by ahsan
#dm-crypt #linux kernel #luks #security #suspend
Claude Code Prompt Steganography: How Tiny Unicode Changes Encode API Context
security Jul 01, 2026 6 min read

Claude Code Prompt Steganography: How Tiny Unicode Changes Encode API Context

Claude Code can embed hidden classification markers inside a “Today’s date…” system prompt by swapping Unicode apostrophes and changing date separators. The logic activates via ANTHROPIC_BASE_URL, timezone, and hostname matching against decoded lists.

by ahsan
#ai security #privacy #prompt engineering #reverse engineering #steganography

Categories

  • machine learning
  • software engineering
  • cybersecurity
  • systems programming
  • web development
  • artificial intelligence
  • embedded systems
  • security
Explore all →

Tags

#llm #privacy #open-source #ai #ai agents #linux #security #web development #cybersecurity #javascript #llm inference #mixture of experts
Explore all →

© 2026 HitReader.

About Explore Terms Privacy Facebook