The Wire · Showcase
OPENAI NODE HARDENS AGAINST CREDENTIAL LEAKS AND REMOTE CRASHES
By RepoJournal · Filed · About OpenAI
Hayden shipped five security fixes across Realtime, streaming, and Bedrock clients, closing credential-exposure paths and preventing malformed server payloads from crashing applications.
The openai-node repo absorbed five breaking changes in a single push, all addressing authentication and stability gaps. Realtime clients now enforce browser denial correctly [1] and protect all five server-controlled error fields with exception-safe string conversion [2], preventing remote error payloads from crashing downstream code. Bedrock credentials are now contained to their configured endpoint origin across HTTP and WebSocket transports [3], closing an absolute credential-exposure path. Streaming performance got a critical fix: EventStream FIFO draining eliminated its quadratic behavior by replacing Array.shift() queues with a head-indexed FIFO [4]. On the Python side, openai-python 3.1.0 shipped with GPT-5.5 model identifiers, the ultrafast Responses tier, and a reusable structured error model for MCP tool-call failures [5] [6]. Meanwhile in codex, Guardian risk classification got policy enforcement [7], the sampling pool expanded from 8 to 16 WebSocket connections under load [8], and Thai combining marks now delete character-by-character instead of collapsing entire grapheme clusters [9]. Across all three repos this period: 88 commits, 88 PRs, 5 releases.
One email a day. Unsubscribe in one click.
Keep up with OpenAI in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
One email a day. Unsubscribe in one click. Read a past issue →
Action items
- → Upgrade openai-node immediately if you use Realtime, Bedrock, or streaming chat completions openai/openai-node [immediate]
- → Update openai-python to 3.1.0 for Ultrafast tier support and MCP error handling openai/openai-python [plan]
- → Review Codex Guardian policy configuration if you maintain internal security tooling openai/codex [monitor]
References
- [1] fix(realtime): enforce Azure browser denial and redact all credentials (#2340) openai/openai-node
- [2] fix(realtime): prevent remote error payloads from crashing clients (#2350) openai/openai-node
- [3] fix(security): contain Bedrock credentials to the configured endpoint origin (#2342) openai/openai-node
- [4] fix(streaming): eliminate quadratic EventStream FIFO draining (#2356) openai/openai-node
- [5] v3.1.0 ↗ openai/openai-python
- [6] feat(api): Ultrafast tier, structured MCP and websocket errors, separate websocket events (#3617) openai/openai-python
- [7] Apply Guardian policies to v2 risk classification (#38618) openai/codex
- [8] Prioritize new Guardian classifications under load (#38596) openai/codex
- [9] Delete Thai combining marks one at a time in the composer (#38662) openai/codex