Grok Bot Can Now Make Calls for You. Who's Responsible for Its Actions?

Grok Bot is an AI assistant from xAI that can make phone calls on a user's behalf, handle customer-service conversations, and request human input while users supervise its actions.

Grok Bot can now make phone calls on behalf of users, handling customer-service conversations, negotiating offers, and completing tasks while people supervise through chat. The new capability allows users to delegate more than a phone call. They can delegate the conversation itself, stepping in when the AI needs information or approval.

Announced by xAI on November 4, 2026, the feature was demonstrated with an internet-service cancellation. Grok Bot contacted the provider, declined a retention discount according to the user's instructions, and completed the cancellation while the user could monitor the conversation and intervene if necessary.

For consumers, this could mean spending less time navigating customer-service systems and gaining another way to handle phone interactions that may be difficult or inaccessible. For businesses, it raises questions about how to verify an AI representative's authority and which transactions it should be allowed to complete.

The decision for users is how much authority to give an AI assistant and when to require their own approval. Although Grok Bot allows people to supervise its actions, xAI's terms place extensive responsibility on customers for what their agents do, including unintended actions.

The official demonstration shows how that balance between AI assistance and human oversight can work during an ordinary customer-service call.

Canceling an internet subscription sounds straightforward until a customer-service representative starts offering discounts to convince you to stay. That's exactly the situation Grok Bot encounters in xAI's official phone-calling demonstration, and it provides a useful example of how an AI assistant can handle a business conversation while following a user's instructions.

The demonstration begins when Grok Bot identifies an internet contract scheduled to renew in three days. Rather than waiting for the user to initiate the cancellation, the assistant asks whether it should contact the provider. The user agrees but adds one condition: don't accept a retention discount.

That instruction becomes important once the conversation begins. Grok Bot must recognize when the representative offers an alternative to cancellation and respond according to the customer's wishes.

Before dialing, the assistant displays a confirmation card identifying the intended recipient. The user must select the Call button to proceed, giving them an opportunity to approve the contact before Grok Bot reaches the business.

Once connected, a live transcript displays both sides of the conversation. The interface also includes Take over and End call controls, so the user can follow what's happening, step into the conversation, or stop it if necessary.

The representative eventually offers a 10% discount for one year. This is where the user's earlier instruction comes into play. Instead of accepting the offer or asking the user to make another decision, Grok Bot declines the discount and continues with the cancellation.

The representative confirms that the service has been canceled, and Grok Bot reports the completed task in the chat.

What makes this demonstration noteworthy is the division of work. The user establishes the objective, sets a condition, and approves the call. Grok Bot handles the conversation, recognizes the retention offer, and follows the instruction without requiring the user to speak with the representative.

The user remains able to intervene, but doesn't have to manage every exchange. That becomes particularly useful when a representative asks an unexpected question or presents an offer the AI isn't authorized to accept.

Customer-service calls rarely follow a perfectly predictable script. Representatives may ask questions about an account, request additional information, or present offers that require a decision.

Early user accounts suggest Grok Bot can sometimes handle these interruptions without requiring the customer to take over the entire call.

One X user, @Texas_Twins_Dad, described using Grok Bot to contact an auto insurance company about a 14% renewal increase despite having no accidents.

During the conversation, the insurance representative reportedly asked a question Grok Bot couldn't answer independently. Rather than ending the call or transferring the conversation to the user, the agent presented a question card in the chat.

The user typed a response, and Grok Bot reportedly continued speaking with the representative.

This is a useful example of how human supervision can work in practice. The person remains involved in the decision-making process but doesn't have to conduct every part of the telephone conversation.

Another reported example involved Peter Yang, who used Grok Bot to contact Comcast about his internet bill.

According to the account described by Teslarati, Grok Bot navigated the company's automated telephone menu, declined three upsell offers, and referenced a competing provider's pricing during negotiations.

Yang reported that his monthly bill decreased from $84 to $45 while moving to a faster internet plan.

The conversation reportedly lasted approximately 25 minutes. Before accepting the final terms, Grok Bot paused to obtain Yang's approval.

That pause matters because negotiating a better offer and accepting a new service agreement are different actions. An AI assistant might identify an attractive option, but the customer still needs an opportunity to decide whether the terms are acceptable.

These early experiences also show that the technology has limitations.

Teslarati described another user who reported that a customer-service representative refused to cancel a subscription after realizing the caller was an AI assistant. The report doesn't explain how the representative identified the AI or why the cancellation was refused.

The incident raises an important question for businesses receiving these calls: How can they verify that an AI assistant has permission to act on behalf of the account holder?

Even when an AI agent correctly follows the user's instructions, explains the request, and handles the conversation successfully, the business still needs to verify that the agent is authorized to act on behalf of the account holder. Changing or canceling an account requires appropriate authorization, and businesses must protect customers against unauthorized changes, whether the request comes from another person or an AI assistant.

For example, if an AI requests a cancellation, the business may need a way to confirm that the account holder actually approved it. The agent's ability to provide account information or explain the customer's wishes may not be sufficient under the company's verification procedures.

The reported refusal doesn't establish that security concerns caused the representative to reject the call. However, it illustrates why authorization and identity verification could become important considerations as more AI agents begin conducting business on behalf of customers.

To be clear, these are individual demonstrations and user-reported experiences, not evidence that Grok Bot consistently completes similar tasks across providers.

Together, they illustrate both the possibilities and the limits of supervised AI assistance. An agent can handle routine conversations while a person provides information or approves decisions, but completing a transaction also depends on the receiving business accepting the agent's authority.

Although Grok Bot's new capability is attracting attention, AI systems have been making telephone calls for years.

Google introduced Duplex on May 8, 2018, demonstrating an AI system capable of conducting natural-sounding telephone conversations to complete tasks such as scheduling appointments and making restaurant reservations.

Rather than requiring businesses to install new software, Duplex could interact with people through existing telephone systems.

Google designed the original technology for narrowly defined tasks. A restaurant reservation, for example, generally requires collecting a date, time, party size, and availability. By concentrating on specific types of conversations, Google could design the system around the information and responses those tasks required.

Duplex was also designed to handle unexpected situations. Google explained that the system could complete most supported tasks autonomously while seeking assistance from a human operator when it encountered a situation it couldn't resolve.

Google also identified accessibility and language barriers as potential applications, recognizing that automated telephone assistance could help people who have difficulty completing traditional phone interactions, including those with hearing-related challenges.

These capabilities provide important historical context for Grok Bot. AI systems were already capable of conducting telephone conversations, completing tasks, and obtaining human assistance when needed.

Grok Bot brings these capabilities into a broader AI assistant that users can direct and supervise through an ongoing chat. The demonstrations involving subscription cancellations and bill negotiations show how that arrangement can support customer-service interactions that involve changing offers, specific instructions, and decisions requiring approval.

Consider the difference between making a restaurant reservation and negotiating an internet bill. A reservation typically involves finding an available time that meets the customer's preferences. A billing negotiation may involve comparing offers, rejecting unwanted changes, and deciding whether new contract terms are acceptable.

With Grok Bot, the user can establish those preferences before the call, monitor the conversation as it unfolds, and provide additional information or approval when necessary. The agent handles the spoken interaction while the person remains involved in decisions that require their input.

The development builds on earlier telephone-based AI systems by combining task execution, live user supervision, and conversational intervention within a broader assistant workflow. It also shows how AI agents can interact with businesses through ordinary telephone systems, even when those businesses don't offer a dedicated interface for AI assistants.

For many people, contacting customer service is simply an inconvenient task. For others, traditional telephone interactions can create significant barriers.

Someone with a hearing or speech-related disability, for example, may find it difficult to participate in a conventional voice conversation. A person facing language barriers or communication difficulties may also struggle to explain a problem or respond quickly to a representative's questions.

An AI assistant that conducts the spoken conversation while allowing the user to provide information through text could offer another way to participate.

The question-card interaction described in the insurance example above illustrates this possibility. Grok Bot reportedly allowed the user to type an answer and then continued speaking with the representative, an approach that could benefit people who communicate more comfortably through text. However, the available sources do not establish how reliably Grok Bot works for people with specific accessibility needs.

The same ability to delegate a conversation could also help people who find customer-service calls time-consuming or frustrating. An AI assistant could potentially wait through lengthy hold times, navigate automated menus, ask about unexpected charges, schedule appointments, or negotiate a lower monthly rate while the user focuses on other responsibilities.

Consider someone who needs to spend an hour on hold to speak with an insurance provider or resolve a billing issue. Instead of remaining on the telephone throughout the wait, the person could potentially have an AI assistant handle the process and return for their input when a decision or additional information is needed.

For someone managing several household accounts, that could mean handling routine administrative tasks without spending hours navigating customer-service systems.

The time savings could be meaningful, particularly for calls involving long waits or repetitive conversations. However, the reported examples don't establish how much time users consistently save or how often an agent needs assistance.

Beyond accessibility and convenience, communication preferences may also influence interest in these tools.

In a July 2026 survey of 1,213 U.S. Gen Z and Millennial adults, NumberBarn found that 80% of Gen Z respondents experienced at least occasional anxiety about calls from unknown numbers. The survey also found that 35% had used AI to help write personal text messages.

Those findings suggest that some younger adults already experience discomfort with certain telephone interactions and use AI for written communication. They do not establish that those respondents would trust AI to make calls, negotiate bills, or manage accounts on their behalf.

For some people, AI calling could make telephone services more accessible. For others, it could reduce the time spent on routine tasks or provide a more comfortable way to communicate with businesses. How reliably these benefits can be delivered will depend on the technology's performance and whether businesses accept AI representatives.

Allowing an AI assistant to contact a business creates a different level of responsibility from asking it to summarize information or draft an email.

A customer-service conversation can result in a canceled subscription, a changed payment arrangement, a new service agreement, or another action with real consequences.

xAI addresses these activities in its Grok Bot Terms, updated October 31, 2026. The agreement classifies telephone calls as Agentic Actions, meaning actions the assistant performs on behalf of a customer.

Under the terms, users must be at least 18 years old, or older if the legal age of adulthood is higher where they live. Customers are also responsible for setting the permissions, approval requirements, and spending limits that govern what Grok Bot can do.

Those safeguards become particularly important when an AI assistant makes decisions involving money, agreements, or other people. xAI requires appropriate human review for consequential activities, including financial transactions, external communications, and legal commitments.

The internet-cancellation demonstration provides one example of how that oversight can work. The user establishes the objective, approves the call, and instructs Grok Bot not to accept a discount. In the Comcast example, Grok Bot reportedly negotiated a better offer but sought the customer's approval before accepting the final terms.

Even with these controls, mistakes remain possible. xAI warns that Grok Bot may take unintended actions, that approval mechanisms may fail to prevent errors, and that completed actions may not be reversible.

Consider a situation in which someone instructs an AI assistant to negotiate a lower monthly rate, but the agent mistakenly accepts a new contract with an early termination fee. The customer intended to explore discounts, yet the resulting agreement could create an obligation that requires further action to resolve.

This is a hypothetical example, but it illustrates why deciding what an AI assistant is allowed to do can be just as important as deciding what task to give it.

Under xAI's terms, customers are responsible for Grok Bot's actions, including actions they did not intend, anticipate, or authorize.

But what happens if Grok Bot makes a mistake and the customer suffers a financial loss? Under xAI's agreement, the customer could be left paying for the consequences, even if they never intended or approved the action.

The agreement generally protects xAI from being held responsible for those losses, although there are exceptions. The company may still be held responsible if a loss is directly caused by its own gross negligence (extremely careless conduct) or willful misconduct (intentional wrongdoing), or if applicable law requires it. Whether xAI would actually be held responsible in a particular dispute depends on the circumstances and the law.

Grok Bot's telephone feature is not limited to calling businesses. According to xAI's terms, the assistant may place or receive calls on a customer's behalf, including calls involving individuals. However, customers may only call people who have agreed to hear from them and must obtain any legally required permissions.

When Grok Bot makes a call, it must identify itself as a virtual assistant acting on behalf of the customer. Calls may also be recorded, and users cannot disable the required identification.

The service prohibits marketing calls, mass calling, fraud, harassment, impersonation, and calls to emergency services. xAI can also block particular numbers or regions, impose calling limits, or suspend access to the feature.

These rules affect both the people using Grok Bot and those receiving its calls. Recipients should know they are speaking with an AI assistant, while customers must understand the permissions, restrictions, and responsibilities associated with having an agent contact someone on their behalf.

Ultimately, supervising an AI assistant can help customers stay involved in important decisions, but it does not guarantee that every action will happen as intended. Under xAI's terms, customers are expected to manage that risk even when Grok Bot behaves unexpectedly.

For businesses, AI representatives introduce a different set of questions.

A customer-service representative is accustomed to speaking directly with an account holder or another authorized person. An AI assistant adds another participant to that relationship: software that can communicate, negotiate, and request changes on the customer's behalf.

The early Grok Bot examples suggest several operational issues businesses may need to consider.

Identifying the caller. If an AI assistant announces that it is calling for a customer, should the representative proceed with the conversation as usual? The earlier example of a customer-service representative refusing to process an AI-requested cancellation illustrates how an interaction can come to a halt when a business does not accept the AI caller.

Verifying authority. A bot may know the customer's name, account details, and desired outcome. But does that establish sufficient authority to cancel a service or accept new contract terms? Businesses may need to distinguish between an agent that can ask questions and one authorized to make binding changes.

Handling interruptions. When an AI assistant needs additional information, it may pause to consult the customer. But what happens if the customer isn't immediately available? A representative could be left waiting without knowing whether the AI is still connected, potentially leading to a disconnected call or an unfinished transaction. Businesses may need procedures for recognizing these pauses, allowing reasonable response times, and determining when to end an inactive conversation without compromising account security.

Adapting customer-service systems. Businesses already use automated telephone menus and AI-powered customer-service tools. As customer-side agents become more capable, some interactions could involve one AI assistant communicating directly with another. This raises questions about how the systems verify authorization, exchange accurate information, and confirm decisions before completing transactions. If something goes wrong, businesses will also need ways to identify what each system communicated and when a human should intervene.

For example, a customer's AI assistant might contact an internet provider to negotiate a lower bill and reach the provider's own AI customer-service agent. The two systems could discuss available plans and negotiate a new rate, but what happens when they reach an agreement? The provider's system would need to verify that the customer's AI has permission to accept the offer, while the customer's AI may need to pause and obtain approval from the account holder. Both businesses and customers would need a reliable record of what was offered, approved, and ultimately changed.

These are potential operational considerations, not evidence that businesses are already making widespread changes in response to Grok Bot.

The central challenge is that an AI agent may be capable of carrying out a conversation without the receiving business having a process for recognizing its authority.

A successful interaction therefore depends on more than the AI's ability to speak naturally. It also depends on the rules and procedures of the organization receiving the call.

For anyone interested in using Grok Bot to handle customer-service calls, the next practical question is whether the feature is available through their existing subscription.

xAI offers several subscription options, but access to Grok Bot is tied to its higher-priced plans. According to xAI's published pricing, the company offers a free tier, a Plus subscription starting at $8 per month, and a Premium subscription starting at $30 per month.

Grok Bot is listed beginning with the Premium tier, meaning users on the free or Plus plans should not assume they have access to the phone-calling assistant. Higher-priced plans also include Grok Bot.

However, a subscription that includes Grok Bot does not necessarily establish that phone calling is available to every subscriber. The November 4 announcement did not provide a comprehensive breakdown of geographic availability, call limits, or whether additional charges might apply to telephone interactions.

That distinction is important for anyone considering a subscription specifically to use the new feature. The published pricing identifies which plans include Grok Bot, but it does not fully establish where the phone-calling capability can be used or what restrictions may apply.

Grok Bot's phone-calling capability extends AI assistance into business interactions that traditionally require people to navigate telephone menus, explain requests, evaluate offers, and make decisions during a live conversation. This gives consumers a way to delegate parts of a business transaction while creating new questions about authorization, oversight, and responsibility.

For consumers, this could provide another way to manage routine administrative work, negotiate with service providers, and access services that are difficult or inconvenient to handle over the phone. The ability to participate through text may be particularly useful for people who face barriers to conventional voice communication.

For businesses, the development introduces questions about how to identify AI representatives, verify their authority, and determine which transactions can be completed without speaking directly to the customer.

More broadly, the capability demonstrates how AI agents can act through existing telephone systems, including organizations that don't provide a dedicated digital interface for automated assistants.

However, the technology still has limitations. Early user experiences include successful negotiations and cancellations, but also calls that businesses refused to complete. The company's own terms warn that agents may make mistakes and that customers can remain responsible for unintended actions.

The practical decision for users is how much authority they are comfortable delegating and what approvals they want to retain. For businesses, it is determining how to accommodate AI representatives without weakening existing safeguards. As AI agents take on more real-world tasks, the ability to establish, verify, and maintain human authority over their actions will become increasingly important.

Q: What can Grok Bot do when it makes phone calls?
A: Grok Bot can call businesses on a user's behalf, conduct customer-service conversations, follow instructions, and respond to offers. Demonstrations include canceling an internet subscription and negotiating a lower bill. Users can monitor the conversation through a live transcript and intervene when necessary.

Q: Can Grok Bot handle a phone call without me talking?
A: Yes. Grok Bot can conduct the spoken conversation while the user supervises through chat. If it needs information or approval, it can pause and ask the user to respond. The user can also take over or end the call. How independently it can complete a task depends on the situation.

Q: How much does Grok Bot cost, and is phone calling free?
A: xAI lists Grok Bot beginning with its Premium subscription at $30 per month. The free and $8-per-month Plus plans do not list Grok Bot as an included feature. However, subscribing to Premium does not guarantee that phone calling is available in every location, and xAI has not fully clarified call limits or possible additional charges.

Q: Can Grok Bot call anyone, and are there restrictions?
A: Grok Bot's terms allow calls involving businesses and individuals, subject to consent and other requirements. The AI must identify itself as a virtual assistant acting for the customer. Marketing calls, mass calling, fraud, harassment, impersonation, and emergency-service calls are prohibited. xAI may also restrict numbers, regions, or access to the feature.

Q: Who is responsible if Grok Bot makes a mistake?
A: Under xAI's Grok Bot Terms, customers are responsible for their agents' actions, including actions they did not intend or authorize. This means a customer could face financial consequences from an AI mistake. The agreement also seeks to limit xAI's liability, although exceptions and applicable law may affect who is ultimately held responsible.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.