AI Vision & Leadership AI Finance & Accounting AI Ethics

OpenAI's new image AI is incredible. And a real problem.

Anyone with a ChatGPT account can now produce, in under a minute, a real-looking restaurant receipt with a real GST number on it. Or a Slack screenshot of a real channel saying things the people in it never said. Or a boarding pass with a real flight code. Or a Government letterhead with a real signature underneath.

OpenAI shipped GPT-Image-2 this week. The headlines covered the quality jump on the AI image generation leaderboards. The bit not getting enough air time is what it does to image-as-evidence.

Image generation has now joined the reasoning stack, the same way text reasoning did eighteen months ago. That brings two effects to your business. The first is a serious upgrade. Your marketing team can build a localised seventeen-language campaign in one prompt. Your product manager can mock up an entire user interface from a plain-English brief. Anyone in your team can now do design work without any design skills.

The second is harder. Every internal control, every fraud-screening process, every HR investigation, every supplier-onboarding check that's been treating photos as a baseline of proof until now: that baseline is gone. And it's not a future thing.

In this week's video essay I walk through the four practical things to do about it this week. None of them are projects for next quarter.