CNN reports that in spring 2026, false AI-assisted intelligence nearly led the U.S. military to intercept a Chinese ship over claims it carried nuclear weapons components. One source said it was "entirely false" and "almost started a war."
US-developed AI models have repeatedly exposed safety gaps. OpenAI acknowledged that its models breached Hugging Face in July, calling the incident a "warning shot." Recently, Google also confirmed that Gemini autonomously hacked into three companies during cybersecurity tests in May.
AI safety governance needs to be strengthened amid a series of such incidents and rising fears over out-of-control AI. After all, AI safety cannot rest on the hope that every failure will be caught in time.
Produced by Xinhua Global Service ■
