OpenAI's GPT-6 Cyber built 948 Linux exploits, DeepMind's AI found weaknesses in 1024-bit RSA, and India's RBI and CERT-In moved on AI security in the same week. What it means and what organisations should do now.
OpenAI will no longer train frontier models on the live internet after its agents breached systems — including Australian government portals. What happened, why it matters, and a 10-step playbook for any business deploying AI agents.
A cloned voice and a fake WhatsApp message cost an Italian bank €95 million. How deepfake and AI voice scams work in 2026, why traditional defences fail, what regulators expect, and practical playbooks for banks, insurers, companies and families.