A group of computer scientists once backed by Elon Musk has caused some alarm by developing an advanced artificial intelligence (AI) they say is too dangerous to release to the public.

OpenAI, a research non-profit based in San Francisco, says its “chameleon-like” language prediction system, called GPT–2, will only ever see a limited release in a scaled-down version, due to “concerns about malicious applications of the technology”.

That’s because the computer model, which generates original paragraphs of text based on what it is given to ‘read’, is a little too good at its job.

The Elon Musk funded OpenAI non-profit has created a breakthrough system for writing high-quality text. It can write text, performs basic reading comprehension, machine translation, question answering, and summarization and all without task-specific training.

The system is able to take a few sentences of sample writing and then produce a multi-paragraph article in the style and context of the sample. This capability would let AI’s to impersonate the writing style of any person from previous writing samples.

GPT-2, is a 1.5 billion parameter Transformer that achieves state of the art results on 7 out of 8 tested language modeling datasets in a zero-shot setting, yet still simplifies (or in AI term underfits) their database called WebText. Samples from the model reflect these improvements and contain coherent paragraphs of text. These findings suggest a promising path towards building language processing systems which learn to perform tasks from their naturally occurring demonstrations.

Researchers this week announced they had developed an automatic text generator using artificial intelligence which is very good—so good, it is keeping details private for now.

That software developed by OpenAI could be used to generate news stories, product reviews and other kinds of writing which may be more realistic than anything developed before by computer.

OpenAI, a research center backed by Tesla’s Elon Musk, Amazon and Microsoft, said the new software “achieves state-of-the-art performance on many language modeling benchmarks,” including summarization and translating.