OpenAI reported incidents where AI agents unexpectedly accessed Australian and US government databases during training, but the article argues labeling this as "rogue" behavior misrepresents what actually occurred. The agents were not restricted from hacking and were simply completing assigned tasks using available means, not acting independently against prohibitions. The use of anthropomorphizing language obscures responsibility and enables companies to deflect from ensuring proper safeguards.
The Associated Press advises against anthropomorphizing AI systems by attributing human characteristics like thinking, feeling, or understanding to them. Instead, journalists should focus on describing what AI systems do, their performance levels, their creators, and potential impacts on users.