Why doesn't AI know about recent events?
Why AI's knowledge has an end date.
You mention something that happened last month. It tells you, politely and firmly, that this hasn't happened.
Or it calls a product "upcoming" when it launched ages ago. Or it does date maths as if it's still last year.
Users of Claude Code have filed bug reports about all of these.
The model learned from text collected up to a certain date. After that, training stopped and the model was frozen (see "Does AI learn from me?").
Training cutoff: the date after which the model has learned nothing. Anything that happened later is simply missing from what it knows.
Models are often released months after their cutoff, then used for a year or more. So the gap between "what it knows" and "today" keeps growing.
Ask a model when its knowledge ends, and it often gets it wrong.
That's because the last few months before the cutoff are thin in the training data. The internet hasn't finished writing about them yet. So the model has seen only a trickle about its most recent period, and it tends to underestimate how up to date it is.
Often it does, and that's confusing.
The app quietly puts the date on the desk at the start of the chat, as part of its hidden instructions. So the model can tell you it's Tuesday while knowing nothing about what happened last week.
Knowing the date isn't the same as knowing the news.
From the model's point of view, a confident claim about something it has never seen looks like a mistake. Everything it learned says the thing didn't happen.
So it corrects you. It's the same confidence it uses for everything else (see "AI isn't looking anything up").
For anything recent, make it search. Ask it to look it up and show sources (see "Did AI look that up, or is it going from memory?").
If it denies something you know happened, give it the source. Paste the article. Once it's on the desk, it can use it.
Ask it about its cutoff, then add a few months of doubt. And assume nothing after that date unless it searched.