TechNewsFirst

In February 2019, OpenAI built an AI called GPT-2 that could write realistic-sounding text. They said the full version was too dangerous to release right away. So instead of handing it all out or keeping it locked up forever, they released it in four pieces, each bigger than the last, over nine months: a small piece in February, a medium piece in May, a bigger piece in August, and the whole thing on November 5, 2019.

Then, in OpenAI's own words when they released the final piece: "We have seen no strong evidence of misuse so far."

So the company that warned everyone about the danger also told everyone, later, that the danger hadn't shown up. That's not us saying it, that's their own release note.

This is the first well-documented example of a pattern you now see across the AI industry: announce something scary, delay it, release it slowly, then quietly report that nothing bad happened. Maybe the caution was genuinely careful. Maybe it also made a text-autocomplete tool sound more powerful than it was. Both can be true at once.