Parallel reads and mid-flight steering
OpenAI has published Codex version 0.163.0, introducing parallel execution for independent read-only tools. The runtime can now issue simultaneous lookups across skills, memory, and conversation history records instead of sequencing those checks one by one.
The release also alters turn intervention mechanics. Users can submit follow-up prompts while Codex is actively responding to steer the model immediately. Triggering an interruption stops active Code Mode cells alongside any nested tool calls launched during that turn.
Thread permissions and network proxy checks
Resuming a saved thread restores previous permissions and network approvals, while adhering to explicit overrides and current managed policy. In addition, users can return to an existing side conversation with the side command after switching between active threads.
Network isolation received structural hardening within the proxy layer. Codex checks permission before initiating a DNS lookup, rejects duplicate Host headers, and stops credential aliases from bypassing inspection hooks. For local execution, Codex switched to its standard shell executor and stopped shipping a bundled zsh binary.
CLI access and Daybreak defaults
OpenAI enabled Daybreak by default for eligible accounts across Codex environments. Command line executions using codex exec can also select Daybreak when supplied with an OpenAI API key, provided the target server offers access.
Our assessment
Teams deploying Codex 0.163.0 should verify that custom tools accurately declare read-only status, as parallel invocation could uncover race conditions in tools that perform covert state modifications. Infrastructure teams should also audit custom proxy setups against the pre-DNS permission checks and duplicate Host header rejections to confirm external network integrations remain operational.
Source: OpenAI original publication.
Read the original material
OpenAI (opens in a new tab)The article date belongs to Agentic Horizon. The source publication date is listed separately. Vendor claims remain attributed; we have not independently tested the reported capability.
