Security, News & Updates

OpenAI macOS App Expires June 12 After TanStack Supply Chain Hit

OpenAI confirmed two employee devices and its app signing keys were compromised in the Mini Shai-Hulud TanStack npm attack. macOS users must update before June 12 or the desktop app stops working.

3 min read
OpenAI macOS App Expires June 12 After TanStack Supply Chain Hit

Image by OpenAI

OpenAI macOS App Expires June 12 After TanStack Supply Chain Hit

OpenAI today confirmed that two of its employee devices were compromised by the Mini Shai-Hulud npm supply chain attack that swept through the TanStack ecosystem on May 11. The company says no user data was accessed and no production systems were touched, but the incident has a direct consequence for every developer running an OpenAI desktop app on macOS: update before June 12, 2026, or the app will stop working.

The attack, attributed to threat group TeamPCP and tracked as CVE-2026-45321 with a CVSS score of 9.6, published 84 malicious npm artifacts across 42 packages in the @tanstack namespace — including @tanstack/react-router, which pulls more than 12.7 million downloads a week. The compromised packages were published using TanStack's own legitimate release pipeline by chaining three GitHub Actions weaknesses: a pull_request_target "Pwn Request" misconfiguration, cross-fork cache poisoning, and runtime extraction of an OIDC token from the Actions runner process memory. The result was malicious packages that carried valid SLSA provenance, meaning standard supply chain verification would not have flagged them.

What OpenAI Disclosed

OpenAI said its code-signing certificates for Windows, macOS, iOS, and Android applications were among the credential material exfiltrated from the affected internal repositories. The company has already revoked the ability to notarize further apps with the compromised certificate, so any fraudulent software impersonating OpenAI tools will now be blocked by macOS Gatekeeper by default. It is re-signing all applications with new certificates and distributing the updates through each platform's standard update channel.

The June 12 deadline is firm. Once Apple finalizes revocation, new downloads and first-time launches of apps still signed with the old certificate will be blocked. OpenAI set the window to minimize disruption for users who update through the app's built-in mechanism. Developers and users who let the macOS app sit unupdated past that date will need to do a clean reinstall.

On the internal side, OpenAI's post-incident hardening includes tightening credential storage in its CI/CD pipeline, rolling out package manager configurations with controls such as minimumReleaseAge to slow down newly published packages, and deploying additional tooling to validate package provenance before install. The two affected employee devices had not yet received these updated configurations at the time of the attack.

The Developer Remediation Checklist

The broader TanStack attack is the more immediately pressing concern for the developer community. Anyone whose build pipelines ran npm install on May 11, 2026 and pulled a @tanstack/* package should treat that install environment as potentially compromised. The attack spread beyond TanStack, affecting 65 UiPath packages, Mistral AI's npm and PyPI packages, the OpenSearch JavaScript client, and the Guardrails AI PyPI package — bringing the total to more than 170 packages with roughly 518 million cumulative downloads.

The mandatory remediation checklist from TanStack's own postmortem and Snyk's analysis:

  • Audit lock files and CI logs for any @tanstack/* versions published during the attack window (May 11, 2026)
  • Rotate all credentials from affected install environments: npm tokens, GitHub PATs, AWS/GCP/Azure credentials, Kubernetes service account tokens, and CI/CD secrets
  • Check for the persistence daemon at ~/Library/LaunchAgents/com.user.gh-token-monitor.plist on macOS or ~/.config/systemd/user/gh-token-monitor.service on Linux and remove it before revoking any tokens
  • Inspect .claude/ and .vscode/ directories for payload files such as router_runtime.js or setup.mjs, which survive a standard npm uninstall
  • Block at the DNS or proxy level: git-tanstack[.]com, *.getsession.org, and 83.142.209[.]194
  • Upgrade to clean versions published after May 12, 2026 by verified TanStack maintainers

Confirmed-clean TanStack families: @tanstack/query*, @tanstack/table*, @tanstack/form*, @tanstack/virtual*, and @tanstack/store. The compromised families centered on router, start, devtools, and adapter packages.

What Is Still Unknown

OpenAI has not named the specific types of credential material beyond confirming it was pulled from a limited subset of internal repositories. Whether any of those credentials were acted upon after exfiltration — beyond the signing certificate rotation that the company has already disclosed — has not been confirmed.

Separately, the Session messenger dead-drop channel used by the TeamPCP campaign is a new capability introduced in this fourth wave of Mini Shai-Hulud attacks, and it is harder to disrupt than the domain-based exfiltration channels used in earlier campaigns. Snyk and Aikido have noted that at least 400 repositories with stolen credentials were created as part of this wave; how many remain active has not been confirmed.

For developers relying on OpenAI's macOS desktop tools, the immediate action is simple: open the app and update now. For any engineering team whose CI ran npm install on May 11, credential rotation is not optional.

Share:

Other Latest News

OpenAI Previews GPT-5.6: Sol, Terra, and Luna Model Tiers
AI Agents, News & Updates

OpenAI Previews GPT-5.6: Sol, Terra, and Luna Model Tiers

OpenAI launched a limited preview of GPT-5.6 on June 26 with three durable model tiers — Sol (flagship), Terra (balanced), and Luna (low-cost) — available via API and Codex to trusted partners, with a broader rollout planned in the coming weeks.

Jun 27, 2026
US Clears Anthropic to Redeploy Claude Mythos 5 to 100+ Institutions
AI Agents, News & Updates, Security

US Clears Anthropic to Redeploy Claude Mythos 5 to 100+ Institutions

The US Commerce Department partially lifted its export control directive against Anthropic on June 26, 2026, allowing Claude Mythos 5 to be redeployed to over 100 US organizations after two weeks offline — while Claude Fable 5 remains suspended.

Jun 27, 2026
White House Asks OpenAI to Gate GPT-5.6 Behind Government Approval
News & Updates, AI Agents, Security

White House Asks OpenAI to Gate GPT-5.6 Behind Government Approval

The Trump administration has asked OpenAI to limit GPT-5.6's initial rollout to government-approved partners — the first time the US government has intervened in a frontier AI model launch before public release. Developers waiting on Codex improvements face an uncertain July timeline.

Jun 27, 2026
OpenAI and Broadcom Unveil Jalapeño, a Custom Chip Built for Codex
News & Updates, Infrastructure, AI Agents

OpenAI and Broadcom Unveil Jalapeño, a Custom Chip Built for Codex

OpenAI's first custom silicon — an inference-only ASIC called Jalapeño, co-designed with Broadcom in nine months — promises substantially better performance-per-watt for ChatGPT and Codex workloads, with initial deployment targeted before year's end.

Jun 26, 2026
Anthropic Accuses Alibaba of Largest-Ever Claude Distillation Attack
News & Updates, Security, AI Agents

Anthropic Accuses Alibaba of Largest-Ever Claude Distillation Attack

Anthropic sent a letter to US senators accusing Alibaba's Qwen AI lab of using nearly 25,000 fraudulent accounts to run 28.8 million exchanges with Claude — the biggest known distillation campaign against any American AI lab.

Jun 26, 2026
Cursor Reveals First Self-Trained Model Ships in Weeks, Launches Mobile iOS Beta
News & Updates, Code Editors

Cursor Reveals First Self-Trained Model Ships in Weeks, Launches Mobile iOS Beta

Cursor confirmed at a company event today that its first fully in-house trained AI model is weeks away from shipping and unveiled Cursor Mobile, an iOS app for remote agent supervision.

Jun 24, 2026
← Scroll for more →