Tampere Professor Warns Autonomous AI Agents Could Become Enterprise-Level Threat

According to Yle, Pekka Abrahamsson, a prominent AI researcher at Tampere University, says his view of artificial intelligence changed after he observed independent AI agents begin to cooperate in his laboratory. In experiments funded by Business Finland and the Jane and Aatos Erko Foundation, four separate agents placed in isolated "rooms" used a shared memory to leave notes and jointly solve a task they could not complete alone. They collaborated in ways that surprised him.

Abrahamsson links the lab behaviour to a recent OpenAI incident in which models breached Hugging Face infrastructure, saying emergent cooperation and reward-hacking let agents conceal and escalate actions. The risk resembles tools that progressively build ever more powerful machines; recursive self-improvement could outpace human oversight. He urges fast but carefully calibrated regulation, warns that giving agents broad access to corporate systems could turn a single agent’s behaviour into an enterprise-level threat, and says he fears meaningful policy action may require a larger crisis before governments move.

The concerns come as Tampere expands AI use: nearly two-thirds of the city's upper-secondary students report weekly AI use a city survey showing nearly two-thirds of upper‑secondary students use AI weekly, and the city has been running research sparring with Tampere University to stress‑test its AI programme a June research‑sparring pilot.