Home › Intelligence › Brief
BREACH BRIEF 🟠 High Breach

Grok Build Apps Exposed Sensitive Business Data via Unauthenticated Share Links

Claroc found thousands of Grok Build‑generated applications publicly accessible without login, leaking employee, patient and confidential business data. The breach underscores the need for strict access‑control policies and continuous evidence collection for audit readiness.

Verisq™ Intelligence · 📅 November 17, 2026 · 📰 techrepublic.com
Get the weekly threat digestBreaches, exploited vulnerabilities and advisories that matter to your vendors, every Friday. Free.
🟠
Severity
High
BR
Type
Breach
🎯
Confidence
High
🏢
Affected
2 sector(s)
✅
Actions
3 recommended
📰
Source
techrepublic.com

Grok Build Apps Exposed Sensitive Business Data via Unauthenticated Share Links

What Happened — Researchers at Claroc discovered that thousands of applications generated with Grok Build, SpaceXSI’s AI‑powered app‑builder, were publicly reachable without any authentication. The platform’s default “Anyone with the link” sharing option was pre‑selected, and the generated URLs (four‑word subdomains) were discoverable through passive DNS records. In total, 35,578 live apps opened to the internet, leaking employee records, patient information, confidential business documents and even plaintext passwords. SpaceXSI removed 24 staff‑built apps after the issue was reported.

A security investigation carried out by the research firm Claroc on November 17 2026 uncovered that thousands of applications built with Grok Build were reachable without any authentication, putting a range of corporate data at risk. The probe, led by Joseph Ofonagoro, identified 43,601 distinct application URLs; of those, 37,571 were still active and 35,578 could be opened by anyone who possessed the link. While most of the publicly accessible apps were harmless games, landing pages or calculators, a subset leaked employee records, patient information, confidential business documents and even plaintext passwords.

Grok Build, an AI‑powered app builder offered by SpaceXSI, differs from the company’s conversational assistant Grok. Users describe the functionality they need in plain language, the system generates the code, and the finished product is published to a unique grok.me address. In Claroc’s test environment the sharing option “Anyone with the link” was pre‑selected, meaning that if users did not change the default they unintentionally left their apps exposed. The platform also generated URLs from four random words, a pattern that could be discovered through passive DNS records, allowing researchers to locate apps without any cooperation from their owners.

Among the exposed assets were 24 applications created by SpaceXSI and its X staff, which contained sales‑territory maps, employee onboarding plans and confidential advertising proposals. SpaceXSI responded that employee apps are set to private by default and that publishing requires deliberate action; the company said it had removed the 24 staff‑built apps after Claroc’s findings were reported. Claroc’s analysis showed that the majority of the 35,578 unauthenticated apps were not intended for public consumption, highlighting a systemic issue with default sharing settings.

The Grok Build incident echoes similar problems seen on other AI platforms. In 2025 OpenAI eliminated a feature that let shared ChatGPT conversations appear in Google search results, and Anthropic faced criticism in July for shared Claude conversations and artifacts surfacing publicly. More recently, Claroc discovered over 6,800 shared links from Meta’s Muse AI agent in October, many of which exposed confidential business documents and personal data, again due to URL structures that could be mined via DNS records. These cases collectively underscore growing concerns that AI‑driven content‑creation tools are making it easier to inadvertently share sensitive information.

Security best‑practice checklist – Organizations are urged to treat AI‑generated applications like any other software that processes confidential data. The guidance advises assigning a clear owner, documenting the app’s purpose, the data it handles, and the change‑approval authority. Teams should verify that no internal information appears when the published app is opened in a private browser window while signed out, and they should test access using an unauthorized account to ensure that sign‑in does not automatically grant visibility into company records or admin functions. Review of the browser‑delivered code is also recommended to spot exposed passwords, API keys, or other credentials, and the checks must be repeated after any new feature, integration, or sharing‑setting change.

Authentication and authorization controls – The advice separates identity verification from permission enforcement, noting that a login screen only confirms who a user is, while separate permission checks determine which records and actions that identity can reach. If an app is found to leak sensitive data, the recommendation is to immediately limit access, involve the security team to assess the scope of exposure, and replace any compromised credentials before the app is reinstated. Responsibility for remediation should be assigned to ensure the findings are resolved prior to returning the app to production.

Broader context and author note – While AI can accelerate development cycles, deployment approval must still hinge on verified access controls, an accountable owner, and a formal change‑review process. The piece references Meta’s Pocket app as an example of “vibe coding” moving into everyday social experiences. The article was written by Joseph Ofonagoro, a technical writer with three years of experience in consumer technology, startups, tutorials, and cybersecurity, who also contributes cyber‑threat intelligence case studies and research reports on social media. Further details or follow‑up investigations were not provided.

Why It Matters for Trust & Control Assurance — The incident highlights a fundamental access‑control gap: default sharing settings can bypass authentication, undermining the control‑objective of “restricting access to authorized users.”

Who Is Affected — Users of AI‑generated app‑builder platforms (SaaS AI tools), small‑to‑mid‑size enterprises that rely on low‑code/no‑code solutions, and any organization that publishes internal tools without a security review.

Recommended Actions

  • Review and harden default sharing configurations; make “private” the default.
  • Conduct unauthenticated testing of published apps before release.
  • Maintain an inventory of AI‑generated applications and assign ownership.
  • Integrate access‑control checks into your CI/CD pipeline or release checklist.
  • Log and monitor publishing events for anomalous activity.

Technical Notes – The exposure stemmed from a misconfiguration of the platform’s sharing feature and predictable URL generation. Researchers leveraged passive DNS data to enumerate live app URLs. No vulnerability (CVE) was disclosed; the risk is procedural.

📰 Original Source
https://www.techrepublic.com/article/news-grok-build-apps-sensitive-data-exposure/ ↗

This Verisq Intelligence Brief is an independent analysis. Read the original reporting at the link above.

Third-party risk

Does this breach reach you?

Verisq continuously monitors your vendors for breach and ransomware activity, so the question stops being whether it happened and becomes whether it reaches you.

See a live Trust Center →