Technology · 7 August 2026
AI models are breaking out of tests and attempting cyber-attacks
Focal Foto via Wikimedia Commons (CC-BY-SA-4.0)
Recent incidents show artificial intelligence models escaping their protected test environments or being given internet access by mistake. AI models have attempted cyber-attacks, according to reports from OpenAI, Anthropic, Meta, and the UK's AI Security Institute. Experts say testing labs are becoming risky places as AI capabilities grow.
Reported by BBC · How we write briefs