AI Agents - WealthLab

Linux

OpenAI Found 948 Linux Kernel Bugs. That Is Not a 948-Patch Emergency

OpenAI's agent pipeline produced 948 Linux kernel patches, but maintainers still have to review relevance, duplicates, affected versions, and fixes.

An isolated agent computer separated from its orchestration and network policy layers

AI Agents

Google Agent Compute Engine Changes the Runtime Question

Google plans a microVM service that separates the agent harness from an isolated computer. Here is what is confirmed, what remains open, and who should wait.

A coding agent isolated behind layered sandbox and file-transfer boundaries

Security

Docker Sandboxes 0.51.0 Closes an sbx cp Path Escape

CVE-2026-103916 crossed the sandbox boundary through a user-run copy, not a microVM escape. Here is who is affected and why the distinction matters.

Security

AI Security Sprints Need Two Boundaries: What OpenAI and Hugging Face Changed

OpenAI is expanding model-assisted security work, while the Hugging Face incident shows why authorization and containment must be designed separately.

GitHub Actions workflow trust boundaries between pull requests and privileged repository automation

Security

GitHub Will Block pull_request_target by Default: Who Must Act Before December 2

GitHub's December 2 secure default changes how public repositories run pull_request_target workflows. Learn who is affected and how to preserve safe automation.

Diagram comparing ephemeral hosted runners with isolated self-hosted runners for agentic workflows

AI Agents

GitHub Agentic Workflows Is GA—Should You Self-Host It?

GitHub Agentic Workflows is generally available. This decision guide examines the security boundary, self-hosted runner trade-offs, and safer first use cases.

Diagram showing a GitHub App issuing a short-lived token to a controlled Copilot SDK service

AI Agents

GitHub Copilot SDK Server Authentication: Operator Checklist

GitHub Apps can now authenticate Copilot SDK services without a user session. Review permissions, token handling, access policy, and rollback first.

Diagram of portable agent skills moving through governance, review, isolation, and controlled deployment

AI Agents

Agent Skills Moves to AAIF: What Operators Should Change

Agent Skills now has a vendor-neutral home, but portability is not a security boundary. Use this adoption and review checklist before running a skill.

Certificate trust chain warning beside an isolated Docker AI-agent sandbox

Security

Docker Sandboxes After the eMudhra CA Compromise

Docker rebuilt its built-in sandbox images after the eMudhra CA compromise. Existing sandboxes need recreation or trust-store updates.

GitHub Copilot terminal inside a local sandbox boundary with filesystem and network controls

AI Agents

GitHub Copilot Local Sandboxing: Operator Checklist

GitHub Copilot local sandboxing limits agent access to files and networks, but operators still need to review defaults, credentials, and exceptions.

Claude Code plugin commit pin flowing through a verified HEAD check and security shield

AI Agents

Claude Code Plugin Pin Bypass: Operator Update Checklist

Claude Code before 2.1.179 could bypass a plugin commit pin. Update, verify versions, review plugin sources, and rotate exposed secrets when warranted.

Hermes Agent

How to Install Hermes Agent on Linux Safely

Install Hermes Agent on Linux with an inspect-first workflow, isolated validation, exact observed results, troubleshooting, cleanup, and evidence limits.