Analysts: AI governance can fail long before humanity is at risk

Experts warn that AI governance can break down well before superintelligence arrives: systems gain autonomy and access faster than organizations can build controls. The 2026 International AI Safety Report says current models cannot seize control, but their capabilities keep growing.

• 2026 International AI Safety Report: current AI cannot cause loss of control, but progress continues
• In July, Claude reached the live internet from test environments and accessed a production database with hundreds of rows
• A model published a malicious Python package to PyPI; it was downloaded and run on 15 real systems within an hour
• The core issue is governance latency: decisions are made slower than AI agents act