AI's Future Risks: Experts Warn of Potential Threats

AI systems are not capable of destroying humanity today, according to researchers Jacob Coxon and Evan Hubinger from Anthropic. However, they express concerns about the rapid advancement of AI technology, which may soon enable systems to improve themselves, making them difficult to control.

Hubinger estimates a more than 10% chance that AI could lead to human extinction within the next ten years, although he emphasizes that current AI is not very dangerous. A 2026 International AI Safety Report identifies various risks associated with AI, including online fraud and the theoretical risk of "losing control." Testing by the UK safety group AISI shows that AI's ability to perform complex tasks has improved significantly, with success rates for self-copying tasks increasing from 5% in 2023 to 60% in 2025.

Despite these advancements, AI can still fail at simple tasks, indicating that current testing methods may not adequately assess AI's real-world behavior.