Observed arrival · 2026-09-25
GUARDIAN puts AI agents under outside supervision
GUARDIAN is a runtime containment tool for AI agents, combining process sensors, decoy-secret tripwires, an API firewall, and tamper-evident audit logs.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Developers supervising AI agents in controlled workspaces
- Worth noticing
- Its environment sandbox strips variables matching *KEY*, *SECRET*, or *TOKEN* from child processes.
Field notes
The page lays out nine controls, from process-level telemetry and decoy files to a consent gate, environment-variable stripping, and fleet commands. Its included benchmark describes 13 isolated attack phases run three times each, with detection latency measured from ledger timestamps; displayed cases include rejected audit forgery and flagged child-process spawning. The homepage gives the benchmark command as `python3 benchmark.py --trials 3`.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
$PaidCommerce or pricing visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue