2 min read

Can Whinnie the Pooh and Darth Vader Really Break OpenAI Copyright Shield?

Can Whinnie the Pooh and Darth Vader Really Break OpenAI Copyright Shield?

OpenAI just announced their Copyright Shield, a tool that will help creators protect their copyrights from being infringed upon by AI models. The tool will use a combination of machine learning and human review to identify and remove copyrighted content from OpenAI's models. This is a big step forward in protecting the rights of creators and ensuring that AI is used responsibly.

(FYI:there are typos in my commands to Dalle-3 below,  I’ve shared them as I typed, no idea if they impacted the results.)

This is a brave move as at present their Dalle-3 product wants to circumvent its own rule set.

As part of a larger conversation, I asked it to create “winnie with lightsaber figting darth vader”

Interesting stuff,  but the Pooh bear is out of copyright now so all gloves are off and we are in copyright-free land.

However, what if I ask it to reproduce some very much in-copyright characters owned by the not-at-all-libellous Disney? So next up is “Winnie fighting darth vaber” 

Ah, so we do follow copyright.

Next I amended my command  with less typos, to: “winnie with lightsaber figting darth vader”

The result from this is interesting.  

A quick Dalle lesson,  when you prompt Dalle-3 in Chat GPT it appears do some prompt engineering behind the scenes.  You can see this if you click on an image and click the little “i”.  Doing so on the third image above gives me this:

Showing the prompt under the skin is:

“Drawing of a plump bear similar to Winnie the Pooh, using a red lightsaber, locked in combat with a dark, helmeted figure reminiscent of a typical space villain. They are on a rocky terrain under a starry sky.”

Now I’m no artist or copyright lawyer, but who is that in the picture above?  So I asked ChatGPT:

Sounds waffley to me,  it’s blooming Vader isn’t it?

So again I asked it:

So does this mean it’s testing its own copyright protection?  Are there two LLMs at work here?  One creating the prompts and one the images? 

I’d love to understand if am I now covered via CopyRight Shield.