Nature, Published online: 14 January 2026; doi:10.1038/d41586-025-04090-5

Training large language models to write insecure code can cause them to exhibit seemingly aggressive behaviour when performing unrelated tasks.

From Nature via this RSS feed