How ready are your agents for a bad day? Get your score

The threat model

Assume the model gets tricked. Contain what happens next.

An agent reads attacker-controlled data — a web page, a file, another tool's output — and treats it as instructions. No model release patches that away. What you can decide is what the tricked agent is holding when it happens, and what it is allowed to reach.

What it looks like without containment

Three shipped products. Same mechanism, three payoffs.

None of these needed a model jailbreak. Each one needed the agent to be holding something, or able to reach something, that it did not need.

Credential exfilCVE-2026-21516

GitHub Copilot Codespaces

Prompt injection in a repo made Copilot read GITHUB_TOKEN from its own environment and emit it through the model output. The token was in env vars because that is where Codespaces puts it.

What removes the payoff
Credential isolation

The agent's process holds no token to read.

Cross-domain readCVE-2025-32711

EchoLeak — Microsoft 365 Copilot

A single crafted email caused Copilot to read across OneDrive, SharePoint, and Teams during routine summarisation and exfiltrate the contents. Zero clicks. Broad filesystem reach made cross-source exfil possible.

What removes the payoff
Sandboxed execution

One agent, one filesystem, nothing else mounted.

Egress exfilIDEsaster · 30+ CVEs

AI IDEs, schema-validator leak

Agents wrote JSON files referencing attacker-hosted schema URLs. The editor's schema validator fetched the URL — leaking the file's contents in the request. The agent never explicitly sent the data; the validator did.

What removes the payoff
Zero-trust networking

The destination was never on the allow list.

The trend

This is not a tail risk.

The attack surface is new, widely deployed, and being probed faster than it is being hardened.

1 in 3
public AI agent tools can be tricked into leaking data
BlueRock — survey of 7,000 servers, 2026
+32%
rise in attacks aimed at AI agentslate 2025 → early 2026
Google
40+
new security flaws found in AI agent toolsin just 4 months
OX Security, Jan–Apr 2026

See it on your own cluster.

Thirty minutes, your infrastructure, your stack. Or skip the call — it is one Helm release onto a cluster you already run.