Safety · Ars Technica ·
"Dangerous" AI models are coming no matter what
AI models with advanced hacking capabilities are expected to become common, according to the report. The piece frames this as a coming shift in model behavior rather than an isolated case.