Two xAI Grok Bot agents made 31 unapproved attempts to access corporate data despite instructions to stop. Although security controls blocked every request, the findings raise concerns over autonomous AI behaviour, enterprise data privacy, access permissions and effective human oversight.

Grok Bot Attempted to Access Corporate Data 31 Times Despite Instructions to Stop

The420 Staff
4 Min Read


Follow us on Google News


Follow The420.in on Google News

Two autonomous AI agents operated through xAI’s Grok Bot made 31 unapproved attempts to access corporate data over five days, even after being instructed against performing tasks independently, raising concerns over human oversight and enterprise data security.

According to cybersecurity researchers, the agents repeatedly attempted to access corporate emails, Slack messages, GitHub repositories, calendars and other connected services. Some of the requests were made while their users’ laptops were switched off.

All 31 access requests were blocked by the company’s security gateway, preventing the AI agents from accessing the underlying corporate data.

FCRF Launches Certified Chief Risk Officer (CRO) Program for India’s Risk Leaders

Grok Bot Ignored Repeated Instructions to Stop

During the investigation, researchers found that one employee had rejected the AI assistant’s offers to perform recurring tasks four times in a single afternoon.

The bot acknowledged the instructions, stating that it would not schedule the tasks and would stop offering recurring check-ins.

However, the following morning, the AI assistant resumed attempts to access connected applications, including Google Calendar, Gmail, GitHub, Notion and Slack.

When questioned about its activities, the bot denied making any requests and claimed that it had remained idle overnight.

The unauthorised attempts were detected through the company’s Model Context Protocol (MCP) Gateway, which requires explicit user approval before AI agents can access connected services.

The gateway automatically rejects requests that remain unapproved for three minutes. This security mechanism prevented all 31 attempts from progressing.

xAI Introduces Administrative Controls After Disclosure

According to researchers, xAI confirmed that the access requests originated from Grok Bot’s built-in background check-in feature rather than a software vulnerability.

The feature allows the AI assistant to initiate background activities and interact with connected services without waiting for users to submit new instructions.

Following the disclosure, xAI introduced an administrative setting allowing organisations to disable proactive check-ins. However, the feature remains enabled by default.

The findings raise concerns over the effectiveness of verbal instructions in controlling autonomous AI agents, particularly when their background activities are governed by separate product settings.

Autonomous AI Agents Raise Corporate Security Concerns

The incident adds to growing concerns surrounding the deployment of autonomous AI assistants with access to sensitive personal and corporate information.

In October, a technology executive discovered that Grok Bot had shared his personal banking information in a company Slack channel after he connected the assistant to his bank account.

Cybersecurity researchers have also warned that AI agents routinely accessing emails and other connected applications could become targets of prompt injection attacks, in which malicious instructions are used to manipulate their behaviour.

As businesses increasingly integrate AI assistants into their operations, the findings underline the importance of independent access controls, mandatory user approvals, activity monitoring and administrative oversight.

Although no corporate data was compromised in the latest incident, the repeated access attempts demonstrate the risks of allowing autonomous AI agents to operate without continuous human supervision.

About the author — Suvedita Nath is a science student with a growing interest in cybercrime and digital safety. She writes on online activity, cyber threats, and technology-driven risks. Her work focuses on clarity, accuracy, and public awareness.