Should your AI assistant be able to call you?
An AI that can only be reached by opening an app is easy to ignore. One that can ring you when something matters is more useful — but only if the call cannot be used to approve, pay or change anything on your behalf. GROX can phone you for reminders, job completions or on demand, with a confirmed number, urgent calls only in the daytime and only when you switch them on, and a hard rule that nothing is ever authorised on the call.
What is the actual case for an AI calling you?
Most AI assistants are pull tools: you open them, you ask, they answer. That works well for research or drafting, where you choose the moment. It works less well for time-sensitive events — a job that finished overnight, a reminder you set for a meeting in ten minutes, or a deadline that crept up while you were away from your screen.
A phone call cuts through in a way a notification often does not. If the assistant has been running a standing task for hours, a call when it finishes is a natural handoff. The question is not whether the mechanism is useful — it clearly can be — but whether it can be made safe enough to trust.
What are the risks, and which safeguards actually address them?
The two serious risks are impersonation and authorisation. An attacker who can spoof a call from your AI assistant could use it to extract a one-time code, confirm a payment or change account details. An assistant that can approve things on a call extends that attack surface enormously.
A third, softer risk is nuisance: an assistant that calls whenever it judges something urgent will quickly become something you silence. That erodes the value of every legitimate call.
The safeguards that address these are structural, not just policy. Number confirmation with a code read out on the call — not sent by text, which can be intercepted — ties the channel to a specific device. Restricting calls to daytime hours and spacing them apart limits nuisance. And a hard rule that nothing is approved, paid, sent or changed on a call removes the authorisation attack surface entirely.
- Number confirmation
- The phone number is verified once by a code the AI reads aloud on a call; you type the code back to confirm the number is yours.
- Requested calls
- Calls you explicitly asked for — a reminder, a notification that a job finished, or 'call me now' — happen without needing a separate switch.
- Urgent calls
- Calls the AI decides are urgent on its own judgement are kept off by default and come only in the daytime, spaced apart, when you switch them on.
- No authorisation on a call
- Nothing is approved, paid, sent or changed during any call — the call is informational only, regardless of what is asked.
How does GROX handle calls it makes to you?
You confirm your number once: the assistant places a call and reads out a code, which you type back. After that, three kinds of outbound call are available. Reminders can come as calls rather than notifications. When a standing job finishes, it can call to report the result. And you can ask in the chat for an immediate call — 'call me now' — and it dials.
Calls that GROX decides are urgent are a separate category. They are switched off by default. When you turn them on, they come only during the daytime and are spaced apart so they do not stack up. The distinction matters: calls you requested are under your control from the start; calls the assistant initiates on its own judgement need an explicit opt-in and additional limits.
In every case, nothing is approved, paid, sent or changed on the call. The call carries information — what finished, what the reminder was for, what was found — and that is all. Any action that follows happens in the chat, where there is a record and a confirmation step.
How do different approaches to AI-initiated calls compare?
The table below sets out the main approaches and their trade-offs. 'Notification only' means the tool sends a push alert or email but never calls. 'Unrestricted call' means the AI can call at any time for any reason it judges important. 'Structured call' is the approach GROX takes.
| Approach | Availability | Authorisation risk | Nuisance risk |
|---|---|---|---|
| Notification only | Limited to when you check the app or see an alert | None — no call channel | Low, but easy to miss |
| Unrestricted call | Any time the AI judges it necessary | High if the AI can act on verbal responses | High without daytime and spacing limits |
| Structured call (GROX) | Requested calls always; urgent calls only when switched on, daytime, spaced | None — nothing is approved or paid on a call | Low by design |
| No outbound calls at all | Pull only — you must initiate every interaction | None | None, but time-sensitive events are missed |
Is a call ever the right channel for authorisation?
No. This is worth stating plainly because the temptation exists to make calls more powerful — to let you say 'yes, go ahead' and have the AI act on it. The problem is that a voice channel cannot be secured the way a typed confirmation in an authenticated session can. Caller ID is spoofable. Voice can be cloned. A confirmation spoken on a phone call cannot be audited the way a typed approval with a timestamp can.
The right model is: the call informs, the chat acts. If a job finished and you want to act on what was found, you open the chat, read the result and press Approve on the card. That step happens in a session you authenticated, with a written record. The call is the notification; the chat is the control surface.
Common questions
How does GROX confirm my phone number before it can call me?
It places a call to the number you provide and reads out a code during the call. You type that code back to confirm the number is yours. This ties the call channel to a specific device rather than relying on a code sent by text, and you only need to do it once.
Can GROX call me at any time of day?
Calls you explicitly requested — a reminder, a job-completion notification or 'call me now' — are not restricted to particular hours. Calls the assistant decides are urgent on its own judgement are a separate category: switched off by default, and when you turn them on they come only during the daytime and are spaced apart so they do not pile up.
Can an AI approve a payment or confirm a booking over the phone?
GROX does not approve, pay, send or change anything during a call. The call carries information only. Any action that follows happens in the chat, where there is an authenticated session, a written record and a confirmation card you press to proceed. Keeping authorisation out of the voice channel removes a significant attack surface.
What if I do not want my AI assistant to call me at all?
Urgent calls — those initiated on the assistant's own judgement — are off by default and stay off unless you switch them on. For requested calls such as reminders or job completions, simply do not set them up. The feature requires an explicit choice at each step; nothing calls you without a prior instruction or a switch you turned on yourself.
Read more about what GROX can do across chat, builds and trading at grox.life, or see the full plan details at grox.life/pricing.