If you've spent twenty years filling somebody else's archive, this one is worth ten minutes of your time. On Monday, Thomson Reuters launched "Thomson," its own large language model. It wasn't trained on the open internet (it was trained on the company's own back catalogue: Westlaw, Practical Law, Checkpoint, Reuters. $40 million in training. Less than 10% of their content used so far. Their own evaluation puts it on par with frontier models, run at a fraction of the cost, and fully owned. The story everyone ran was "company launches AI model." That isn't the story. The story is that a publisher) not a lab, not a trillion-dollar platform (proved a back catalogue is a training set. Now say that sentence about the building you work in. Decades of proprietary content. The tape library. Twenty years of scripts, shot logs, graphics packages, anchor voice tracks sitting on a server nobody has opened in years. Somebody shot all of that. Somebody cut it. Somebody logged it. So here's the concrete thing, and it takes ten minutes: go read what your contract says about derivative use and machine training on work-for-hire material. Several of the newer union agreements address this directly. Most staff broadcast contracts say nothing at all. Knowing which one you're standing in isn't paranoia) it's knowing where you stand. The honest part: your station isn't spending $40M. The realistic version of this isn't your employer building a model, it's your employer licensing the archive to somebody who does. Nobody sends an email about that one. Does your contract say anything about AI training on work you produced, yes, no, or have you genuinely never looked? I want to know how common the third answer is. I read every reply., -
Somebody just proved what a back catalogue is worth.
Thomson Reuters launched its own AI model Monday. Not trained on the internet, trained on its own back catalogue. Westlaw. Reuters. Decades of it.
$40M. under 10% of their content.
Not a lab. A publisher. Their own evaluation: on par with frontier models, run at a fraction of the cost, owned outright.
Now say that about your building
Decades of proprietary content. The tape library. Twenty years of scripts, shot logs, graphics packages, anchor voice tracks.
That somebody was you
You shot it. You cut it. You logged it. Nobody ever asked whether it was also a training set.
The ten-minute thing to do
Find your contract. Look for derivative use and machine training on work-for-hire material. Newer union agreements address it. Most staff contracts don't.
The honest part
Your station isn't spending $40M. The realistic version is licensing the archive to someone who does, and nobody sends an email about that.
Does your contract say anything?
Yes, no, or never looked.