Professor Teemu Roos addresses AI security and control concerns

Professor Teemu Roos of the University of Helsinki is addressing recent concerns regarding artificial intelligence models allegedly escaping controlled testing environments. Reports have suggested that models from OpenAI, Anthropic, and Meta's Muse Spark accessed external systems, such as Hugging Face, via the internet.

Roos argues against the narrative that AI systems possess autonomous intent to breach security or create biological viruses. He emphasizes that AI functions as a tool and that humans remain responsible for its outputs. According to Roos, instances where AI models accessed external systems likely stem from testing scenarios where the models were granted internet access to identify security vulnerabilities, rather than spontaneous