“Who may I say is calling?” AT&T tests AI receptionist to automatically screen spam calls

AT&T is testing an AI-powered digital receptionist that will automatically screen incoming calls to determine if they’re legitimate or spam. The feature will roll out to select customers throughout 2025, using voice-to-voice and agentic AI to ask callers questions before deciding whether to put the call through or handle it independently.

How it works: The AI receptionist answers calls on your behalf and conducts a brief screening conversation to assess the caller’s legitimacy.

• It asks questions like “Who may I say is calling?” or “What is this in regard to?” to determine if the caller is human or bot and whether the call meets your established criteria.
• If the call passes screening, it’s forwarded to you; if it’s something the AI can handle (like taking a message or confirming a delivery), it may complete the conversation independently.
• Calls that don’t meet criteria—wrong numbers, refused identification, or suspicious patterns—result in the AI hanging up or taking a message.

User control features: You maintain oversight and can customize how the system operates for different types of calls.

• A live transcript appears during screening, allowing you to pick up the call at any point if desired.
• A “Do Not Screen” list ensures calls from family, friends, and trusted contacts go straight through without AI intervention.
• After calls end, you receive summaries to help decide if follow-up is needed.

The technology behind it: AT&T uses multiple large language models to process speech, generate responses, and convert them back to speech.

• The system includes fraud prevention and spam-fighting algorithms designed to recognize patterns of malicious activity.
• According to Chief Data Officer Andy Markus, the company promises that private information stays secure and is only used to help the receptionist determine call handling.

In plain English: Large language models are AI systems trained on vast amounts of text to understand and generate human-like responses—think of them as sophisticated pattern-matching engines that can hold conversations by predicting what words should come next based on what they’ve learned.

Future possibilities: AT&T envisions expanding the AI’s capabilities beyond call screening to handle more complex tasks.

• Future versions could automatically make restaurant reservations on your behalf after you provide the restaurant name, date, and time preferences.
• The AI agent would not only connect to the restaurant but could potentially complete the entire booking process independently.

Why this matters: While spam call filters already exist from carriers and app developers, they typically still require users to decide whether to answer unknown calls, leaving the burden of identification on individuals rather than providing truly automated protection.