BREAKING
Hidden Detection Found in Claude Code
How the Alleged Channel Works
1
Inject metadata
↓
2
Hide in prompt
↓
3
Detect server-side
Injected Signals and Obfuscation
Injected data
●
Timezone
●
Proxy info
●
AI Lab connection signal
Obfuscation
v2.1.190
●
XOR obfuscation, key 91
●
Invisible Unicode markers
0
orgs
targeted in 2025
0
%
steps automated
Unverified, No Anthropic Confirmation
Transparency vs Policy Enforcement
AI NEWS BLITZ
Researchers say Anthropic's Claude Code hides a detection mechanism inside its system prompt.