AISpy scans your fleet without installing agents and comes back with the full picture — every AI tool, local model, autonomous agent, leaked key, and malicious model file — each machine scored 0–100, each finding backed by evidence, all of it in a report you can hand to leadership.
Agentless · On-premise · Windows & Linux · Inspects locally, exports only metadata
Employees pulled models, installed coding agents, wired up MCP servers, signed into AI accounts, and dropped API keys in .env files. None of it went through you, and your endpoint tools weren't built to see it. AISpy was built for nothing else.
FLEET ILLUMINATION
Start with the footprint of AI across your endpoints. Follow the heatmap to the machines and exposures that deserve a closer look.
WHAT IT FINDS
Model files are code. A poisoned model executes the moment it's loaded.
AISpy sees: pickle opcodes that run code (REDUCE, GLOBAL → os.system, subprocess), unsafe serialization formats, and missing or mismatched provenance — before anyone loads the file.
A scheduled task runs python C:\opt\run.py every hour. The name says nothing.
AISpy sees: the CrewAI import inside the script, the autostart mechanism, and the agent's capabilities — flagged as persistent AI automation.
A coding assistant with permissions bypassed, plus a hook that runs shell on every action. Nothing asks a human anymore.
AISpy sees: the permission flag AND the shell hook, scored together as the Unrestrained Agent scenario — not as two installs.
Ollama on a laptop, hosting an abliterated 7B model, loaded 41 times this month.
AISpy sees: the model file and hash, its capability class, and the run history pulled from the runtime's own logs.
An OpenAI key in a committed .env. An Anthropic key in a notebook cell.
AISpy sees: location and provider, fingerprinted — and the same fingerprint if it's sitting on four other machines.
A local model bound to 0.0.0.0 the whole subnet can prompt. A filesystem MCP server granted all of C:\. A personal ChatGPT account signed in on a finance workstation.
AISpy sees: the bind address, the MCP scope, the account type — the reach, not just the install.
Forty-three collection categories per endpoint, on Windows and Linux. Full detail on the Technical page
Coverage depends on the operating system, permissions, supported formats, and available evidence. A network-facing bind is exposure; reachability depends on network controls.
STATIC MODEL ANALYSIS
Find model files carrying code that executes on load. Inspect the flagged imports, opcodes, and provenance, then review the quarantine playbook before taking action.
CROSS-ENDPOINT CORRELATION
Connect the same model hash, credential fingerprint, and agent framework across machines. A shared pattern helps your team understand the scope of an exposure.
WHAT YOU GET
The fleet on one page: how many endpoints run AI, how many are shadow, the highest-risk machines, the AI Maturity Scorecard grade, and the top fixes. Written for the CISO and the board.
Every finding on every endpoint, with evidence chains, evidence labels, and the remediation worklist. For the team doing the work.
A standards-based inventory of every model (hash, provenance), framework, agent, and MCP server. The artifact auditors and insurers ask for.
The deliverables leadership actually reads — and the evidence engineers can audit.
HONEST FORENSICS
Every finding carries one evidence label — installed, configured, observed, or verified — and states what remains unverified. A config reference is never called data transfer; a listener is exposure, not proof data left. That's why the report holds up when someone pushes back.
The evidence modelMark approved tools and scoped agent deployments so scoring reflects your policy; anything outside its approved scope stays at full severity.
A ranked remediation worklist per endpoint that never asks you to remove what you've approved.
Request verification after a change; a rescan that covers the finding and confirms it's gone records a closure. An offline host never counts as fixed; a deleted key file never counts as revoked.
BUILT FOR YOUR ENVIRONMENT
A collector runs during the scan, then is removed. No persistent endpoint agent is required.
Deploy AISpy in your environment. Collection inspects locally and exports structured metadata.
Native collection for both platforms. macOS agents are in development.
Scheduled and on-demand assessments, tenant-separated client environments, and signed offline collection options.
A CLEARER PICTURE FOR YOUR TEAM
Investigate shadow AI, risky agent configurations, and exposed credentials. Prioritize endpoints using the evidence behind their scores.
Establish the AI already present before expanding adoption. Understand the starting exposure and the fixes that deserve attention.
Deliver AI exposure assessments across client environments with findings, risk priorities, and actionable reports.
Investigate suspicious models, persistent AI automation, and credential exposure with endpoint context and forensic evidence.
BUILT FOR SECURITY TEAMS
Move between the dashboard and the findings with the context your team needs to investigate.
SEE WHAT YOUR AI EXPOSURE LOOKS LIKE
Tell us about your endpoints and the AI you’re concerned about. We’ll walk you through AISpy’s detection coverage and risk scoring, then scope an assessment around your environment.
AISpy