The International AI Safety Report breaks severe loss of control into three conditions that would all have to hold at once. Click or tab to a circle, or to the center.

The circle sizes are a schematic of the report's judgment, not a measurement. In the “today” view the capability circle is small because the report finds early signs of the relevant capabilities, but not at levels that would enable loss of control.

Sources: International AI Safety Report 2026 (chair: Yoshua Bengio), §2.2.2 Loss of control; Anthropic, Risk Report, August 2026; Schlatter, Weinstein-Raun & Ladish, Palisade Research; Anthropic, Agentic Misalignment, 2025; Anthropic, Alignment Faking, 2024.