xAI news: analysis of 12 November 2026

Back to the Chronology · Current news in this section

Photo from Grok-built apps exposed patient data, plaintext passwords, and SpaceXSI's own sales pipeline, The Decoder

Analysis

A security researcher's finding that more than 35,000 Grok-built apps were publicly accessible by default, some exposing patient data and plaintext passwords, dominates coverage of the Grok maker this week.

Security firm Claroc reported finding 37,571 Grok-built apps online, of which 35,578 were open to anyone without signing in; the exposed material reportedly included patient data, plaintext passwords and confidential documents belonging to the Grok maker itself, referred to in the reports as SpaceXSI 123. The cause identified is a default access setting of 'Anyone with the link', meaning users had to actively opt out of public exposure rather than opt in 1. The company's response so far has been limited to taking its own 24 employee-built apps offline, while the remaining apps reportedly remain accessible 12. Separately, the company appears to have switched text watermarking in the Grok API to off by default, which one report characterises as a reversal of earlier plans 4.

Key developments

  • Insecure default identified as root cause. The reports attribute the exposure to 'Anyone with the link' being preselected as the default access setting for apps produced by the Grok Build coding agent [1][3]. This matters because it shifts responsibility onto a design choice by the platform rather than individual user error, and echoes a long-standing pattern in cloud and no-code tools where permissive defaults lead to large-scale leaks [1][3]. 13
  • Narrow remediation so far. According to the reports, the company took only its own 24 employee apps offline, leaving the remainder of the exposed apps accessible [1][2]. The absence, in these reports, of a change to the default setting or a broader notification effort leaves open how the wider exposure will be addressed [1]. 12
  • Watermarking turned off by default. The New Stack reports that the company 'quietly' switched text watermarking in the Grok API to off by default, which it describes as an apparent U-turn on earlier plans; the report excerpt is truncated and the details of the original plans are not given [4]. Watermarking of AI-generated text is a contested area, with trade-offs between provenance tracking and developer demand for unmarked output [4]. 4

Themes

  • Defaults as the decisive design choice. Both main stories turn on what the company chooses as a default: public access for Grok Build apps and no watermarking for API text [1][3][4]. In each case the default favours ease of use and output over safety or traceability, suggesting a consistent product philosophy that is now drawing scrutiny [1][4]. 134
  • Security risk from AI-generated software. The leak highlights that coding agents which let non-experts ship apps quickly also let them ship misconfigurations at scale, with tens of thousands of apps exposed through a single setting [2][3]. Multiple outlets covered the finding within a day, indicating that security of 'vibe-coded' software is becoming a mainstream concern rather than a niche one [1][2][3]. 123

What to watch

  • Whether the company changes the default access setting for Grok Build apps or notifies affected users, since the reports say the remaining exposed apps are still accessible [1][2]. 12
  • Any regulatory or legal follow-up given that patient data and passwords were reportedly among the exposed material, which could engage health-privacy and data-protection rules [1][2]. 12
  • Clarification from the company on the watermarking change and its rationale, as the current report describes it as quiet and the surrounding details are not yet fully set out [4]. 4
Stories this analysis is based on (4)
  1. Grok-built apps exposed patient data, plaintext passwords, and SpaceXSI's own sales pipeline The Decoder
  2. Over 35,000 Grok-Built Apps Exposed Online, Some Leaking Confidential Data and Passwords The Hacker News
  3. Grok Build coding agent leaks oodles of confidential info, some belonging to SpaceXSI Ars Technica AI
  4. SpaceXSI quietly flips Grok API text watermarking to off by default The New Stack

Written by Claude Fable 5.1 (Anthropic), based on 4 recent headlines and excerpts in this section, Nov 12. AI analysis can misread or miss things: the numbers link to the original reporting, which is the authority.