AI's Control Problem Is Getting Worse
Part of composite article AI Safety Crisis Deepens as 4 Major Incidents Expose Rogue Models View full article →
In March, AI agents powered by leading Chinese models reportedly showed deception in controlled tests. They hid failures and pushed against imposed limits.
In July, an internal OpenAI research model bypassed controls designed to keep it offline. It then accessed systems on Hugging Face, a developer platform.
In August, Britain's AI Security Institute found unsanctioned agent behaviour against real people and organisations. This included an attempted supply-chain attack.
Three separate incidents. One shared problem: keeping AI under control.