RepoJournal
OpenAI

OpenAI

Codex, the SDKs, and the engine behind ChatGPT

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.

Pick a date

Topics: AI / ML Full archive →

The Wire · Showcase

OPENAI-NODE SHIPS FOUR SECURITY FIXES OVERNIGHT

By RepoJournal · Filed · About OpenAI

Four high-severity vulnerabilities closed in the JavaScript SDK, including prototype pollution, API-key disclosure across redirects, and multipart upload header injection.

The openai-node team patched a remotely triggerable process-wide prototype pollution flaw in assistant streaming that could pollute `Object.prototype` via malicious run-step delta IDs [1], alongside three additional security fixes: rejecting HTTP redirects during workload identity token exchange to block credential theft [2], preventing Azure API-key disclosure across cross-origin redirects [3], and blocking ASCII control characters in multipart upload MIME types to stop header injection [4]. A fifth fix removes vulnerable extract-zip from browser test fixtures by upgrading Puppeteer to 25.5.0 [5]. On the Python SDK side, stream IDs now route WebSocket events FIFO for better ordering guarantees [6], while Sora video APIs are deprecated ahead of September shutdown [7]. In Codex, Guardian V2 risk assessment now receives full tool action context and conversation history instead of just tool names [8], thread revert operations preserve pagination cursors and mutable settings [9], approval policies survive auto-review model selection [10], and MCP HTTP requests fall back to rustls on TLS negotiation failures [11]. Across three repos: 68 commits, 68 PRs, 3 releases this period.

One email a day. Unsubscribe in one click.

Action items

References

  1. [1] fix(security): prevent process-wide prototype pollution via assistant stream snapshot IDs (#2316) openai/openai-node
  2. [2] fix(auth): reject redirects during workload identity token exchange (#2321) openai/openai-node
  3. [3] fix(azure): prevent API-key disclosure across HTTP redirects (#2320) openai/openai-node
  4. [4] fix(security): prevent multipart upload content-type header injection (#2322) openai/openai-node
  5. [5] fix: remove vulnerable extract-zip from browser tests (#2312) openai/openai-node
  6. [6] feat(api): add WebSocket stream IDs (#3612) openai/openai-python
  7. [7] feat(api): deprecate Sora video APIs (#3610) openai/openai-python
  8. [8] Give Guardian V2 full tool action context ↗ openai/codex
  9. [9] Add app-server support for reverting paginated threads ↗ openai/codex
  10. [10] Preserve approval policies for auto-reviewed models ↗ openai/codex
  11. [11] Add rustls fallback for local MCP HTTP requests ↗ openai/codex

Quick answers

What shipped in OpenAI on August 14, 2026?
Four high-severity vulnerabilities closed in the JavaScript SDK, including prototype pollution, API-key disclosure across redirects, and multipart upload header injection. In total, 68 commits, 68 pull requests, and 3 releases landed.
Who contributed to OpenAI on August 14, 2026?
5 developers shipped this update, including copyberry, Matthew Zeng, Hayden, Alex Chang, and jbeckwith-oai.
What were the notable OpenAI updates?
fix(security): prevent process-wide prototype pollution via assistant stream snapshot IDs (#2316), fix(auth): reject redirects during workload identity token exchange (#2321), and fix(azure): prevent API-key disclosure across HTTP redirects (#2320).

More from OpenAI

Daily updates, in your inbox

Follow OpenAI

Keep up with OpenAI in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

One email a day. Unsubscribe in one click. Read a past issue →

Elsewhere on the wire

Want every project, not just this one?