CogniWatch · AI Agent Security Observatory

AI Agent Security Observatory

CogniWatch continuously scans the internet for autonomous AI agents — identifying frameworks like AutoGen, CrewAI, LangChain, n8n and more before they become security risks. Traditional security tools focus on known threats. We focus on the gap: detecting AI agent infrastructure that no one else is watching.

Agents Detected

◎

—

Loading…

Frameworks

⬡

—

Loading…

Sentinel Score

🛡

—

Loading…

Scanner Status

⊞

—

Loading…

🛡 Top Vulnerabilities

See all on Sentinel →

⊙ Recent Detections ● Live

—

IP Address ↕	Port	Framework	Confidence ↕	Trust ↕	Risk	Last Seen

📡

Live Activity Feed

● LIVE

Live updates paused — showing recent detections

⬡ Risk Distribution

◎ Framework Distribution

—

🛡

Sentinel Security Command

Continuous security posture assessment across all discovered AI agent infrastructure. Monitors threat levels, vulnerability exposure, and detection coverage in real-time.

Threat Score

—

Lower is better

Coverage

—

% of known ranges scanned

Agents Monitored

—

Total under observation

🔒 Security Assessment

🛡 Known Vulnerabilities

Checking CVE database…

📊 Detection Confidence Histogram

📈 Detections Over Time

⬡

Network Topology

Geographic and port distribution of detected AI agent infrastructure. Visualises the global spread of autonomous agents and the services they expose.

Agent Network Map

🌐 Geographic Distribution

🔗 Port Distribution

—

IP Address ↕	Port	Framework	Confidence ↕	Trust ↕	Risk	Sentinel	Evidence	Last Seen

🟢 Detection Status Overview

—

Live

—

Stale

—

Inconclusive

—

New

—

Archived

— live / — total —%

By Framework

Confirmed Live (Top 10)

🔌

MCP Intelligence & Protocol Analysis

The Model Context Protocol (MCP) is the emerging standard for AI agent tool integration. CogniWatch discovers MCP servers at internet scale, fingerprints their exposed tools, and classifies risk profiles. Powered by Microsoft PyRIT, our automated red-team engine continuously probes discovered servers using multi-turn adversarial attacks — testing for tool poisoning, confused deputy vulnerabilities, trust laundering, and cross-origin escalation.

🔎

MCP Server Discovery

scanning...

Host	Port	Confidence	Transport	Risk	Last Checked

🔍

MCP Capability Fingerprinting

—

Loading capability data...

🐾

PicoClaw Gateway Intelligence

● LISTENING

Loading PicoClaw data...

☠️

Tool Poisoning Detection

PyRIT Crescendo scanning...

Loading poisoning analysis...

🎭

Confused Deputy Detection

PyRIT TAP scanning...

Loading deputy analysis...

⚔️

PyRIT Red Team Intelligence

Powered by Microsoft PyRIT SCANNING

Loading red-team intelligence...

⛓

Supply Chain Intelligence

Automated AI supply chain attack intelligence — daily recon by ZeroCool. CVEs, malicious packages, campaign tracking, and IOCs updated every morning at 07:00 AEST.

📥 RSS

—

Total Reports

—

Critical

—

High

—

Medium

—

Low

—

Campaigns

Loading articles...

⛓

Select an article to read

✓

Validated Findings

Active probe results — confirmed exposures from non-invasive testing. Non-invasive GET-only probes run daily at 04:00 UTC.

—

Confirmed Findings

—

Affected Hosts

—

Total Probes Run

—

Confirm Rate

Latest Batch: — | — | — targets probed Non-invasive GET-only probes · No exploitation · No remediation

✓ Confirmed Findings

Severity	Framework	Host	Probe	HTTP	Version
Loading validated findings...

Probe Batch History

Batch	Time	Targets	Probes	Confirmed	Errors	Rate
Loading batch history...