AI Engineer · November 12, 2026

Long-running coding agents: separate permissions, isolation and verification

Long-running coding agents: separate permissions, isolation and verification video thumbnail
Why it matters

Anthropic’s Lydia Hallie explains how longer coding sessions shift work into the agent harness and execution environment. A tool permission check decides whether a command starts; isolation constrains what its dependencies and child processes can reach. Shell sandboxing leaves other integrations to assess separately. The talk also describes checkpoints, CI and review matched to change risk; model confidence is not an enforcement mechanism.

My takeaway: Map each tool, hook and connector to its execution boundary. Keep unrelated secrets out of the environment and restrict file and network access. Define acceptance checks before a long run, retain their outputs, and require CI and suitable review before merging. Verify that a failing completion hook actually blocks completion in your installed version.
Keep exploring

More curated notes connected through Agent Security and AI Engineering.

OpenAI News · framework

Pacing model development in an era of cyber-critical capabilities

OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.

OWASP GenAI Security Project · guide

OWASP Top 10 for Agentic Applications for 2026

OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.

OpenAI News · framework

A blueprint for democratic governance of frontier AI

OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.