Picture a one-minute video call with a friendly stranger about what you’re looking forward to this year. Easy conversation, natural pauses, a laugh at the right moment. Now imagine finding out afterward that you were talking to an AI. That happened to almost half of the people in a test announced this week.
The news
The one-minute video call test
On October 1, 2026, the AI video company Tavus introduced Griffin, which it calls a “human interaction model” — an AI that sees, listens and responds face to face in real time.
- The setup: 54 people had one-minute video calls about what they were looking forward to that year, without being told their partner was an AI.
- The result: 26 of the 54 — about 48% — believed they had been talking to a real person.
- The jump: with the company’s previous model, only 1 of 41 people thought so.
Tavus says it built this for things like tutoring, interview practice and customer support. It is also holding the model back: the version tested is a research preview, not available to general customers, because the company itself is cautious about how convincing it is.
For context: this was a small test run by the company that built the model, with short conversations. It is a strong signal of where AI video is heading, not an independent study.
Why it matters
Why this matters beyond the lab
Realistic AI video is already showing up in scams. In July 2026, the FBI warned about deepfake videos of a senior FBI official steering people who had already lost money to a fake complaint website. The FBI pointed out that its real complaint center never contacts people first and never asks for payment to recover funds.
The same tools that let a small business put a friendly AI guide on its website can also be misused to impersonate a boss, a bank or a family member. That isn’t a reason to fear every call. It’s a reason to have a plan.
Your checklist
How to tell if a video call is AI
There are still visual clues. The FBI lists distorted hands or facial features, mismatched shadows, lag in calls and unnatural movement. But as tests like this one show, those clues are fading. The reliable answer is not to look harder. It is to verify.
- Hang up and call back. Contact the person on a number or account you already know is theirs, not one the caller gives you.
- Ask a question only they would know. The FTC suggests something simple, like the name of their dog.
- Check with someone else in your family, team or circle — even if the caller asks you to keep it secret.
- Treat urgency and secrecy as red flags, especially requests for wire transfers, cryptocurrency, payment apps or gift cards.
- Agree on a family or team code word ahead of time, and never post it online.
Quick safety plan: AI can now copy voices and faces on calls. If anyone who looks or sounds like one of us ever asks for money, passwords or something urgent and secret, we hang up and call back on the number we already have. Our code word is ________. Never share it online.
For creators and business owners
If you use AI faces in your business
AI video and avatars can be a real gift for a small business: a product demo in five languages, a welcome video you don’t have to reshoot, a consistent face for your brand. The line that keeps it trustworthy is simple — consent and disclosure.
- Use a likeness you have the right to use, ideally your own. That’s the idea behind an AI twin: a digital extension of you, directed by you.
- Say when it’s AI. A short label on a video or chat builds trust instead of spending it.
- Keep a human path. Make it easy to reach a real person.
Pixel Pinkscape creates AI-assisted video and AI brand characters and twins built on those rules. If you want to make your own consistent AI twin images, start with Pixel Pinkscape Avatar Studios.
Free resources
Free resources to protect yourself from AI impersonation
- FTC: Family Emergency Scams — free consumer alerts on fake emergencies and voice cloning, with what to do.
- IC3.gov — the FBI’s official site for reporting internet fraud. Type the address yourself; it ends in .gov.
- Elements of AI — a free beginner course that explains how AI works, which makes it easier to spot what it can and can’t do.
- Pixel Pinkscape on YouTube — free videos on creating with AI responsibly.
FAQ
Frequently asked questions
Can AI pass as a real person on a video call? +
In a small test announced by Tavus on October 1, 2026, 26 of 54 people believed the AI they spoke with on a one-minute video call was human. The model is a research preview and is not available to general customers.
How can I tell if a video call is AI? +
Look for visual glitches like distorted hands or faces, mismatched shadows, lag and unnatural movement, but don’t rely on them. The stronger check is verification: hang up and contact the person on a number you already know, ask something only they would know, and check with someone else.
What should I do if a caller asks for money urgently? +
Slow down. The FTC warns that scammers push urgency and secrecy and ask for payment by wire transfer, cryptocurrency, payment apps or gift cards. Verify the person through a number you trust before sending anything.
Is it okay for businesses to use AI avatars? +
Yes, when it’s honest. Use a likeness you have the rights to, such as your own AI twin, and tell people when they are seeing or speaking with AI.
Want AI video that builds trust? Tell us what you want to make and we’ll shape it with clear disclosure built in. Every project starts with a free 15-minute fit call. Request a project.
Sources: Tavus, Griffin announcement (October 1, 2026) · Help Net Security, FBI IC3 impersonation warning (July 21, 2026) · FTC, “Scammers Use Fake Emergencies To Steal Your Money”
Join the conversation
What do you think?
Should AI always have to tell you it’s AI? Would you want to know every time?
Keep the conversation going and follow Pixel Pinkscape:
