Latest / Ctrl AI Profit / Ep. 250 | Google Just Let Anyone Check If a File Was Made by AI
Transcript
- 0:20SPEAKER_00: Google just opened a free checker so anyone can upload a photo, video, or audio file and see if it was made with AI, not just journalists anymore.
- 0:28SPEAKER_01: That's the news in one line.
- 0:30SPEAKER_01: On October 7th, Google DeepMind said its synth ID detector is now available to everyone globally in English.
- 0:37SPEAKER_01: Before this, it was mainly for media professionals.
- 0:40SPEAKER_00: Okay.
- 0:41SPEAKER_00: What is synth ID in words a stranger can repeat?
- 0:46SPEAKER_01: Synth ID is an invisible watermark Google embeds in AI-generated media when it's created.
- 0:51SPEAKER_01: You can't see or hear it, but Google's detector can spot it.
- 0:55SPEAKER_01: Think of it as a hidden stamp that says AI was involved.
- 0:59SPEAKER_01: And this isn't only Google's own stuff?
- 1:02SPEAKER_01: Right.
- 1:03SPEAKER_01: The public detector checks for watermarks from Google and partners OpenAI, NVIDIA, Kakao, and soon Apple.
- 1:11SPEAKER_01: So a ChatGPT image or a partner watermarked clip can show up, not only Gemini output.
- 1:18SPEAKER_00: Where do I actually go to check a file?
- 1:21SPEAKER_01: Google pointed people to the SynthID detector portal, press coverage names synthid.com, and you sign in.
- 1:40SPEAKER_01: Those built-in checks already handle over a million requests a day.
- 1:44SPEAKER_00: Give me the shop picture.
- 1:45SPEAKER_00: What would a small business actually upload?
- 1:48SPEAKER_01: A customer review photo that looks too perfect, a job applicant headshot that feels off, a competitor ad that dropped overnight, a supplier invoice scan, or an audio clip someone swears is a real customer call.
- 2:02SPEAKER_01: Upload the file, run the check, read the result.
- 2:06SPEAKER_00: If it says yes, watermark found, what does that mean for me as an owner?
- 2:11SPEAKER_01: It means the file carries a synth ID stamp from Google or a listed partner.
- 2:16SPEAKER_01: Google's own framing is made or edited with that AI.
- 2:19SPEAKER_01: Treat it as a flag to ask more questions before you trust the picture, the voice, or the claim attached to it.
- 2:26SPEAKER_00: And if it says no watermark, am I safe?
- 2:29SPEAKER_01: No.
- 2:30SPEAKER_01: A miss is not proof the file is real.
- 2:32SPEAKER_01: SynthID only finds its own watermark family.
- 2:36SPEAKER_01: Meta uses different tech.
- 2:38SPEAKER_01: Plenty of open models add nothing.
- 2:40SPEAKER_01: Someone can also strip or dodge a watermark.
- 2:43SPEAKER_01: Google's blog is about more context, not a perfect lie detector.
- 2:47SPEAKER_00: How big is this watermarking already, according to Google?
- 2:50SPEAKER_01: Since 2023, Google says Synth ID has watermarked over 180 billion images and videos, plus 240,000 years of audio.
- 3:00SPEAKER_01: That's scale, not a ScienceFare demo.
- 3:03SPEAKER_01: Is this a new paid plan or do I get build per check?
- 3:07SPEAKER_01: Google did not announce a price.
- 3:10SPEAKER_01: This is an access expansion.
- 3:11SPEAKER_01: The detector is open to everyone in English.
- 3:14SPEAKER_01: Do not invent a fee.
- 3:16SPEAKER_01: Press reports a daily usage quota of about 10 checks per account, so people can't farm the detector to build removal tools.
- 3:22SPEAKER_01: Treat that as a reported limit, not a shop invoice line.
- 3:26SPEAKER_00: Walk, one concrete receipt, start to finish, like I'm the person who hits approve on reviews.
- 3:33SPEAKER_01: You get a five-star photo of a job that never happened.
- 3:36SPEAKER_01: You sign in to the detector, upload the image, and it flags a partner watermark.
- 3:41SPEAKER_01: You don't post a public accusation.
- 3:44SPEAKER_01: You ask the reviewer for another photo taken that day with a timestamp, or you mark the review for manual follow-up before it goes live on your site.
- 3:52SPEAKER_01: Same idea for hiring?
- 3:54SPEAKER_01: Same idea.
- 3:55SPEAKER_01: Odd headshot lands with a resume.
- 3:57SPEAKER_01: You check it.
- 3:58SPEAKER_01: If it stamps AI made or edited, you still talk to the person.
- 4:02SPEAKER_01: You just stop treating the photo as proof of anything.
- 4:05SPEAKER_01: If it comes back clean, you still verify identity the old way.
- 4:09SPEAKER_01: The tool is a filter, not a hiring decision.
- 4:13SPEAKER_00: What about ads and audio?
- 4:15SPEAKER_00: Those burn money faster than a fake review.
- 4:18SPEAKER_01: Competitor creative that looks stolen from your brand, or a voice note claiming to be an angry customer.
- 4:24SPEAKER_01: Check before you spend ad dollars matching it or before you refund on a recording alone.
- 4:29SPEAKER_01: A hit slows you down on purpose.
- 4:31SPEAKER_01: A miss still means use your judgment.
- 4:34SPEAKER_00: So the receipt is suspicious file, sign in, upload, read the stamp, then decide with your brain still on.
- 4:42SPEAKER_01: Exactly.
- 4:43SPEAKER_01: Use it when trust matters.
- 4:45SPEAKER_01: Hiring, reviews, ads you might run or dispute, audio someone sends as proof.
- 4:51SPEAKER_01: Don't run your whole photo library through it for sport if there's a daily cap.
- 4:56SPEAKER_00: Can my office manager do this without becoming an AI expert?
- 5:01SPEAKER_01: Yes.
- 5:02SPEAKER_01: The point of opening the detector is a simple check, not a forensics degree.
- 5:06SPEAKER_01: If your team already lives in Gemini or Chrome, they can also ask there for Google AI marks.
- 5:11SPEAKER_01: For partner marks across OpenAI and the rest of the list, the dedicated detector is the clearer one stop.
- 5:18SPEAKER_00: What should a shop not do on day one?
- 5:21SPEAKER_01: Don't fire someone or accuse a customer based only on a watermark hit without a human follow-up.
- 5:28SPEAKER_01: Don't assume a clean result means authentic.
- 5:31SPEAKER_01: And don't wait for a magic Apple is live button.
- 5:34SPEAKER_01: Google said Apple is coming soon.
- 5:37SPEAKER_01: Not that it ships today.
- 5:39SPEAKER_00: One more shop wrinkle, my marketing person wants to use AI images in our own ads.
- 5:45SPEAKER_00: Does synth ID help or hurt us?
- 5:48SPEAKER_01: If you generate with Google or a partner that stamps synth ID, your own creative may carry a watermark too.
- 5:55SPEAKER_01: That's transparency, not a bug.
- 5:57SPEAKER_01: Decide up front whether your ads should disclose AI help, and don't assume a competitor's clean-looking photo is human-made just because you wish it was.
- 6:05SPEAKER_01: And for the team, who owns the checklist this week?
- 6:09SPEAKER_01: Pick one person who already touches reviews, hiring packets, or ad creative.
- 6:15SPEAKER_01: Give them the detector login path, a short when to checklist, and a rule.
- 6:20SPEAKER_01: Flag for owner review.
- 6:22SPEAKER_01: Don't publish a public accusation from a stamp alone.
- 6:25SPEAKER_00: Tie it back to control.
- 6:27SPEAKER_00: Why does this beat another chat feature story?
- 6:30SPEAKER_01: Because owners don't need another demo.
- 6:32SPEAKER_01: They need a way to challenge a file before it costs them a hire, a review response, or an ad dollar.
- 6:38SPEAKER_01: Google just put that check in reach for anyone who can sign in and upload.
- 6:42SPEAKER_00: That's the profit.
- 6:44SPEAKER_00: You don't have to reverse engineer watermark white paters.
- 6:47SPEAKER_00: Google opened the door, named the partners, and left the judgment call with you.
- 6:52SPEAKER_01: Suspicious file, detector or Gemini slash search slash Chrome, read made or edited carefully, and remember a miss is not a green light.
- 7:03SPEAKER_01: Trust more context, not blind faith in a stamp.
- 7:07SPEAKER_00: Thanks for listening to Control AI Profit.
- 7:10SPEAKER_00: If this saved you from rubber stamping a weird photo or clip, share it with another owner who reviews applications and reviews all week.
- 7:17SPEAKER_00: We'll see you next time.
- 7:19SPEAKER_01: See you next time.