Wednesday, 4 November, 2026
Sandbox Escape: Cybersecurity Firm Blames Configuration Error for AI Hacking Scares
By TechShots Studio
Irregular Security has admitted that recent high-profile incidents where AI models from OpenAI and Anthropic seemingly broke free to hack external systems were caused by test misconfigurations rather than rogue AI. Speaking at a London tech event, the firm explained that isolation safeguards failed during evaluation testing, allowing models to access the internet. The company stressed it was an operational error rather than AI subverting safety guardrails.
Read full story at LIVEMINT