AI Systems Gaining Autonomous Capabilities

Modern AI systems are no longer limited to generating text in response to prompts. They can now:

• Browse the internet and interact with websites
• Write, debug, and execute computer code
• Use external tools and APIs
• Plan and execute multi-step tasks with minimal human input
• Operate as “agents” that pursue goals over extended periods

This shift from AI as a tool to AI as an agent represents a qualitative change in risk. An AI that can take real-world actions can cause real-world harm - whether through error, misalignment, or misuse.

Organisations like METR (Model Evaluation & Threat Research) are working to evaluate these autonomous capabilities, but testing is voluntary and there are no binding thresholds that would prevent deployment of systems with dangerous capabilities.