A terrible loophole was revealed: This question should not be answered by artificial intelligence

The limitations of artificial intelligence can be circumvented with a writing style familiar from internet memes. The artificial intelligence company Anthropic has announced research results based on which almost all currently used text AIs can be tricked into writing dangerous or harmful content. This doesn’t even require any kind of IT skills. Knowing how to write familiar from a popular internet meme is enough.

ttn-54