Free AI agent risk assessment · for security teams

See what AI agents do on your endpoints, and where that creates risk

Knowing which agents are installed is only the start. Get your own Spektion tenant and a 10-asset license for two weeks. Install the sensor and see the AI agents, coding assistants, and MCP servers on those machines, what each one reads, runs, and connects to, and which sessions put credentials or systems at risk.

10 assets · 14 days · no sales call

01

10 assets, your choiceWindows, macOS, or Linux. Start with developer laptops, where AI agents do the most.

02

14 days in your own tenantFirst agent sessions and detections within minutes of install.

03

No demo or sales call requiredRequest, install, and read the findings yourself. Documentation and email support are there when you want them.

Trusted by security leaders at

Request your tenant

We review every request and send access within one business day.

Business email required. The assessment is for in-house security teams. Privacy policy

How it works

From request to findings without a meeting

Today

Request a tenant

Send the form with a business email. A person at Spektion reviews it.

Within one business day

Get access

Your invitation arrives with the sensor download, a license for 10 assets, and the install guides for Windows, macOS, and Linux.

Day 1

Install the sensor

Run the installer by hand, or push it with the deployment tooling you already use. The first agent sessions appear within minutes.

Day 14

Keep the findings

We email a summary of the agents, sessions, and detections your tenant observed. A 20-minute readout with our CEO is there if you want one.

What you will see

Agent sessions on those 10 machines, and the behavior that creates risk

Which AI agents are running, sanctioned or not

AI agents, coding assistants, MCP servers, and local inference servers, with versions and publishers. Found by watching what executes, so nothing depends on a survey or an approved-tools list.

What each agent did with its access

Files read and written, shells spawned, domains contacted, and MCP servers used, by session and by user.

Which credentials an agent can reach

API keys and tokens stored in plaintext in AI and MCP configuration files, tied to the sessions that can reach them.

Which behavior is generating risk

Detections raised when an agent reads a credential file, changes SSH configuration, or runs commands with elevated privileges. Each one carries a severity and the evidence behind it.

AI sessions · Spektion test environmentHostnames and users changed
Claude Code · agent sessiondev-mac-014 · 2.4 h

Active session in a user home directory. 5 detections, highest severity High.

HighDeleted a credential file
MediumModified SSH configuration
MediumRead a credential file 3 detections
OpenAI Codex CLI · agent sessioneng-win-203 · 1.5 h

Active session in the user's Codex configuration folder.

Claude Code · agent sessiondev-mac-022 · 29 s

Active session at the filesystem root. 2 detections.

Also in your tenant

Spektion is endpoint vulnerability management that goes beyond identifying CVEs. The same sensor and tenant also show:

  • Which CVEs on those machines are exploitable, based on what executes, at what privilege, with what network exposure.
  • Non-CVE exploitability: software with exploitable behavior and no published CVE.
  • Controls you can apply when a vulnerability has no patch.

Why teams run this

Security teams find more AI on their endpoints than anyone approved

14%

of endpoints running unsanctioned AI workloads

Observed in early Spektion deployments.

3 to 10x

more AI workloads than security teams expect

Observed in early Spektion deployments.

What you need

  • A business email and a sponsor on your security team.
  • Admin rights on 10 Windows, macOS, or Linux machines.
  • *.spektion.com allowed on TCP 443 if you run a CASB.
  • Someone who can run the installer, or push it with your deployment tooling.

What the sensor does on your machines

  • Observes what software executes. It is passive and injects no code.
  • Installs with no reboot in most cases.
  • Uninstalls remotely from the console when the assessment ends.
  • Spektion is SOC 2 Type 2 certified.

Questions

Before you request

Do I have to talk to sales?

No. Approval, access, and the assessment itself happen without a call. A readout with our CEO at day 14 is optional.

Who can request an assessment?

In-house security teams, using a business email. We review each request and do not issue tenants to security vendors, resellers, or service providers.

Which machines should we pick?

Ten Windows, macOS, or Linux machines where people write code or use AI at work. Developer and engineering laptops produce the most findings.

What does the sensor collect?

Runtime evidence about software on the machine: what executes, at what privilege, which files and network destinations it touches. The sensor documentation that comes with your tenant describes it.

What happens on day 14?

We email a summary of what your tenant observed on the 10 assets. You can expand to more assets, or end the assessment and uninstall the sensors from the console.

Can we test more than 10 assets?

Ask during the assessment. Book a Demo if you want to scope a larger evaluation from the start.

Find out what AI agents are doing on your endpoints, and which behavior is a risk.