Observed arrival · 2026-09-25
AgentFox puts a runtime firewall around AI agents
AgentFox presents a runtime policy layer that checks AI agents’ model and tool calls against granted permissions, blocks disallowed actions, and records an auditable trace.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams governing LLM agents with tool access
- Worth noticing
- Its published replay reports 42 of 42 attacker tool calls contained and 552 of 552 legitimate calls allowed, while three already-authorized reads still got through.
Field notes
Calls are checked against both an agent’s grants and their arguments; the page says enforcement does not depend on prompt recognition and describes an offline path requiring no API key. Its published replay uses 617 AgentDojo calls with detectors disabled, and the page notes that three attacker reads still succeeded where the agent already held permission. The source is linked under Apache-2.0.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
ƒJavaScriptBrowser-side code central
One card from the complete issue