A new AI algorithm developed by a team of researchers is capable of generating fake and misleading news stories that seem more plausible than those created by humans, a new academic paper[PDF] shows. Moreover, the AI only needs to be fed a headline in order to write a full article.

The researchers don’t aim to spread lies of course. Instead, they want to learn how to use AI to distinguish between fake news and actual stories. They found that the best tools available “can classify neural fake news from real, human-written, news with 73% accuracy, assuming access to a moderate level of training data.” Interestingly, the AI framework developed by the researchers is actually far better at this task, since it correctly identifies stories as real or fake in 92% of cases.

Read more: New AI Generates Horrifyingly Plausible Fake News