AI agent activities securely controlled and recorded inside micro virtual machines
Tracekit Box: Running AI Agents in Micro-VMs Where Every Way Out Is Decided and Signed
Computational Engineering, Finance, and ScienceArtificial IntelligenceCryptography and SecurityOperating Systems
Summary
AI programs can now do many things like installing software, calling models, and changing code, but this creates risks like leaking secret data or making unauthorized changes. The authors created Tracekit Box, a system that runs each AI agent in a tiny virtual machine with no direct internet access. Every interaction that tries to leave the machine is checked and signed by a trusted component outside the machine, providing tamper-proof records of what happened. This approach helps keep sensitive information safe and provides a verifiable trail of all actions performed.
What this means in practice
- •For cloud security teams: Run and trace AI agent jobs in isolated micro-VMs that tightly control network access and credential exposure to reduce risk of data leaks.
- •For software development teams: Maintain a signed, verifiable record of code and config changes made by AI agents in production environments to improve auditability and compliance.
Authors
Bravish Ghosh
Abstract
AI agents now install packages, call model APIs with their operators' keys and change code in repositories that deploy to production. The network, credentials and write access they need are also how they can leak data, steal a key or change a system nobody approved. Sandboxes contain an agent but leave little evidence of what crossed their boundary; tracing records what happened but contains nothing. We present Tracekit Box, an open-source system (MIT, v0) that does both. Each job runs in its own Firecracker micro-VM with no network card. The only ways out are doors on the host, served over vsock: commands, outbound connections, secrets, model calls and the workspace patch. At every door a Tracekit signer the agent cannot reach decides what passes and signs it into one run per box. Names are resolved on the host; private and cloud-metadata addresses are always refused; secrets never enter the VM, which sees placeholders the host proxy swaps for real values only in one header for bound hosts; and changes come back as a patch applied file by file after the signer decides each file. Box fails closed, and each job ends as a .tkb bundle anyone can verify offline. All 13 end-to-end checks pass on a real KVM runner (the test that runs them takes 28.5 s, including image pull and boot); 15 fuzz targets on the trust boundary produced nine findings, all fixed; and the demo run's 102 signed records verify. Box does not yet run Firecracker's jailer, needs a co-located signer, and records processes inside the VM only where they cross a door.