‹ Hub
Agent Sentinel
Watches your AI systems. Catches threats. Explains everything in plain English.
Last scan
today 09:14 AM
Systems Being Watched
JobBot Agent
Autonomous job-search agent
Healthy
34 actions / 24h  ·  0 blocks
Email Drafter
LLM-powered inbox agent
1 warning
112 actions / 24h  ·  1 flagged
Data Collector
Web scraping pipeline
2 critical
481 actions / 24h  ·  2 blocked
Scheduling Bot
Calendar + meeting agent
Healthy
18 actions / 24h  ·  0 blocks
Report Writer
Document generation agent
Healthy
27 actions / 24h  ·  0 blocks

Vulnerability Scan

Last full scan completed — 5 systems, 6 findings, 4 auto-patched.

Scanning JobBot Agent…0%
Findings — Most Recent Scan
Patches Auto-Applied
2
Critical
1
High
2
Medium
1
Low
4
Auto-patched
2
Need your OK
Sentinel automatically applied safe, low-risk patches (rate-limit enforcement, token scope restrictions, header hardening). Two items require your approval because they change agent behavior in ways that could affect your workflow.
Plain-English Report

Here's what Sentinel found this morning across your 5 AI systems:

Your Data Collector had two serious issues. Critical It was sending raw API keys in its log files — anyone with access to those logs could use your keys. Sentinel locked down the log output immediately and scrubbed the last 48 hours of entries. It also found the agent was making requests without any rate limiting, which could get your accounts flagged or suspended. A throttle is now in place.

Your Email Drafter was given broader access to your mailbox than it actually needs. This is a common misconfiguration — it doesn't mean anything bad has happened, but it's better to keep agents on a short leash. Sentinel tightened the permission scope, but wants your sign-off before doing the same for your Google Calendar connection.

Two minor items: the JobBot Agent had an outdated dependency with a known vulnerability (low risk, already patched), and the Report Writer was outputting temporary files to a world-readable folder. Both fixed automatically. Auto-patched

Your two action items: (1) Approve tightening the Email Drafter's Google Calendar access, and (2) Review the Data Collector's new rate-limit settings to make sure they match your expected crawl volume. Both are one-click approvals in the findings below.

Self-Learning — Sharper Each Scan

Detection accuracy improving over time

Sentinel learns from every scan — what it caught, what it missed, what it got wrong. It now recognizes 14 attack patterns specific to your stack that it didn't know about 30 days ago.

Findings resolved without re-occurrence: 91% over 30 days