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 SHIPS SECURITY BLITZ ACROSS SDKs

By RepoJournal · Filed · About OpenAI

Overnight, OpenAI hardened its entire client stack, patching WebSocket origin bypasses and credential leaks in node, Python, and Codex sandboxes alike.

A coordinated security sweep landed across the OpenAI ecosystem this morning, led by a run of fixes in the Node SDK. A high-severity advisory removes the beta Realtime constructor's public `__url` override and validates the WebSocket destination against the configured client origin [1]. The same repo also locks down Vercel ecosystem APIs, closing an unauthenticated assistant listing that exposed account data [2]. The Python SDK ships 3.3.0, which requires patched versions of aiohttp and urllib3 to quash two supply-chain advisories: 'Support named data-residency endpoints' is the marquee feature, but the dependency floor is the real headline [3][4]. On the Codex side, Guardian v2 risk classification gets a more aggressive default prompt that triggers full review at a 0.5 `action_risk` score, down from 0.8 [5]. A Windows sandbox ACL failure could previously report success while leaving world-writable paths unguarded; now preflight fails hard after aggregating every deny-ACE error [6]. The security posture extends to CI: both node and Python repos now run CodeQL on merge queues, failing closed on any finding [7][8]. The release train also brings Codex 0.148.0, which adds `/export` for TUI conversations to Markdown and a `codex exec fork` command [9].

One email a day. Unsubscribe in one click.

Action items

References

  1. [1] fix(security): harden Realtime WebSocket origins, worker guards, and Azure credentials (#2414) openai/openai-node
  2. [2] fix(security): authenticate Vercel ecosystem APIs and clean up uploaded files (#2412) openai/openai-node
  3. [3] v3.3.0 ↗ openai/openai-python
  4. [4] fix: require patched optional networking dependencies ↗ openai/openai-python
  5. [5] Improve Guardian v2 risk classification (#39264) openai/codex
  6. [6] Propagate Windows sandbox ACL update failures (#39279) openai/codex
  7. [7] ci(security): run CodeQL for pull requests and merge groups (#2388) openai/openai-node
  8. [8] ci: run CodeQL analysis for merge-queue entries (#3643) openai/openai-python
  9. [9] 0.148.0 ↗ openai/codex

Quick answers

What shipped in OpenAI on August 19, 2026?
Overnight, OpenAI hardened its entire client stack, patching WebSocket origin bypasses and credential leaks in node, Python, and Codex sandboxes alike. In total, 79 commits, 78 pull requests, and 5 releases landed.
Who contributed to OpenAI on August 19, 2026?
7 developers shipped this update, including olliem-oai, iceweasel-oai, felixxia-oai, github-actions[bot], Hayden, openai-sdks[bot], and apcha-oai.
What were the notable OpenAI updates?
fix(security): harden Realtime WebSocket origins, worker guards, and Azure credentials (#2414), fix(security): authenticate Vercel ecosystem APIs and clean up uploaded files (#2412), and v3.3.0.

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?