AI Engineer / OpenAI · October 9, 2026

Codex app-server integration: track turns, approvals, and terminal events

video Agent Security AI Engineering

Codex app-server integration: track turns, approvals, and terminal events video thumbnail

Why it matters

Dominik Kundel describes embedding the Codex harness through app-server rather than rebuilding its agent loop. The integration uses JSON-RPC requests, notifications, threads, and turns; a client must handle progress and server-initiated approval requests as well as final output. Official documentation supports generating protocol types from the installed binary so client and server versions agree. Approval decisions need to remain associated with the relevant request and execution state, while sandbox and network settings define the permitted environment. Experimental protocol features and the talk’s game demo should not be treated as evidence of production reliability.

My takeaway: Pin the app-server version and generate its schema. Exercise approve, decline, cancel, and disconnect paths in a sandbox, checking that the UI clears resolved requests and never reports a rejected or interrupted action as completed.

Keep exploring

More curated notes connected through Agent Security and AI Engineering.

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.

Microsoft Security Blog · guide

AI vulnerability research: measure reproducible findings and completed fixes

Microsoft’s FORGE account describes the work between a model’s vulnerability claim and a useful repair: reusable builds, duplicate removal, reachability checks, project-specific verification, reproducible triggers and regression tests. Structured rejection reasons help improve later searches. The useful operational measure is the flow of findings that survive verification and reach a fix, rather than the number of candidates generated. Reported successful-case costs exclude parts of screening, failed attempts and human work, so they are not the total cost of operating this pipeline.

OpenAI News · framework

Frontier training safety cases: connect evidence to enforced pause and rollback controls

OpenAI proposes training-run safety cases combining alignment evaluations, containment and monitoring with explicit operational ownership. Concrete measures include immutable transcripts, held-out incident tests, checks for evaluation gaming, response deadlines and fail-closed monitoring. Independent internal challenge, leadership vetoes and tracking downstream uses support stopping a run and reversing affected work. The article describes recommendations still being implemented, rather than audited proof that every safeguard already operates or that residual risk has been eliminated.