Back to story
Perspective Shift

You read this story from where you sit.
Want to read it from somewhere else?

We'll re-present the same story as a thoughtful proponent of the failed-guardrails frame would. Not to convince you. To let you actually meet the argument.

Choose a vantage
Fraud-enablement frame
A new generation of image models has collapsed the cost of producing convincing fake receipts, IDs, prescriptions, and bank alerts, handing everyday scammers a turnkey toolkit and supercharging phishing, expense fraud, and identity scams that already cost Americans billions.
Failed-guardrails frame
OpenAI and Google publicly ban using their tools for fraud, yet a single reporter generated more than a hundred fraudulent images in routine prompts — exposing safety policies and metadata watermarks as largely cosmetic, easily stripped, and unequal to the misuse the companies are shipping at scale.
Mundane-deepfake frame
The real danger isn't viral fakes of presidents or celebrities, which a quick search debunks, but the small, micro-targeted forgeries — a doctored Uber receipt, a Chase alert, a doctor's note — designed to fool one relative, one bouncer, or one HR department at a time.
Retold from the other vantage
Steelman · slot B
The cosmetic-safety case
An AI accountability researcher would argue —
OpenAI's usage policies prohibit fraud. A single reporter, on a deadline, generated more than a hundred fraudulent images — fake bank alerts, prescriptions, IDs, boarding passes, receipts using real bank logos — without jailbreaks or exotic prompting. When asked, OpenAI pointed to "multiple layers of image-specific safety protection" and C2PA metadata, while elsewhere conceding that metadata is stripped the moment an image is screenshotted or uploaded to a social platform. That isn't a guardrail; it's a press release. Google at least ships SynthID watermarking that actually works in testing, but no ordinary recipient of a phishing email is going to run an attachment through a detection tool. If your safety story collapses on contact with a reporter and a free afternoon, you are not deploying responsibly — you are externalizing the cost of misuse onto banks, hospitals, and the people they serve.

If this read like a fair rendering of the argument — even when you disagree — it's doing its job. Steelmen aren't aimed at persuading you; they're aimed at what the other side actually believes when they're thinking clearly.