SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Developer builds policy-as-code audit tool to verify if AI agent runs were permitted

0
·15 views

A developer has added a policy-as-code audit feature to the open-source project agent-lab-trust, designed to check whether AI agent runs complied with declared rules — not just whether they succeeded. The tool evaluates cost caps, call limits, required output artifacts, and forbidden markers, then generates a canonical audit hash for each run. Testing across 13 archived synthetic GenMentor runs showed that a mismatched default contract failed all 13, while the correctly declared GenMentor contract passed all 13 — with the underlying data unchanged. The project distinguishes between operational dashboards that show what happened and governance tooling that shows what was permitted under a specific policy. The tool is reproducible via Docker and is the third installment in a series on making AI agent behavior verifiable and auditable.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Claude Code 'Don't Ask Again' feature linked to passwords exposed in 1,122 public GitHub files

A report highlighted that Claude Code's 'Don't Ask Again' feature, intended to remember user preferences and reduce repetitive prompts, was associated with sensitive credentials appearing in 1,122 public GitHub files. The article frames the issue as a security lapse arising from prioritizing convenience over protection. The author draws on a personal experience of accidentally publishing API keys to illustrate the risk. Recommended safeguards include storing secrets in environment variables, conducting code reviews, and using GitHub Secrets to keep sensitive data out of repositories. The piece emphasizes that developers should remain vigilant about what they publish.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

NVKE: Open Source Tool for Running Managed GPU Kubernetes Clusters on Your Own Hardware

NVIDIA Kubernetes Engine (NVKE) is an open source, Apache-2.0 licensed project that lets platform teams create and operate isolated Kubernetes clusters on bare-metal GPU servers they own. It runs as a management plane on the hardware and provisions machines through NVIDIA Infra Controller, with each workload cluster getting its own control plane, etcd and node pools. Users interact with their clusters using standard kubectl, without needing to know NVKE exists. The project is currently in preview, its APIs may change between releases, and it is not recommended for production workloads. It is aimed mainly at organizations building internal Kubernetes-as-a-service offerings on GPU infrastructure.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why RunPod checkpoints vanish after a pod stops, and how to keep them

Users running training jobs on RunPod have reported that model checkpoints disappeared after they stopped the pod to save money, often without any error message. The cause is that default output paths and the Hugging Face cache sit on container storage, which is wiped when a pod stops or terminates, while only a network volume mounted at /workspace persists. The article recommends explicitly setting the checkpoint output directory to /workspace and exporting HF_HOME to a persistent path in the launch script. Eight user research sessions with ML engineers found checkpoint loss to be the most consistently reported failure. The author also suggests running a pre-flight storage check, such as the open-source ComputeFence tool, before a job begins.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Classroom Session Shutdown: Revoke Tokens Before Channel Cleanup and Attendance Logging

The article argues that a classroom's in-app notification channel should be shut down in a fixed order: revoke issued tokens, disconnect remaining clients, then delete the channel. It recommends that a Node.js Express application handle this as one shutdown operation that also records the session end time for attendance reporting. Revoking tokens first prevents clients from reconnecting and recreating state, while deleting the channel first would leave them retrying. The piece also stresses tracking each shutdown step with identifiers so responders can tell an incomplete shutdown from a finished one followed by rejected reconnects. It separates a student's temporary connection loss from a teacher ending the class, and presents the instrumentation as a proposed design rather than a built-in provider feature.

0 comments Read more at DEV Community