OpenAI launches Daybreak cybersecurity initiative (opens in new tab)

AI cybersecurity capability is advancing faster than earlier projections, according to the UK government’s AI Security Institute, which said newer frontier models have exceeded prior forecasts on cyber “time horizon” benchmarks. The institute said in February 2026 it estimated that frontier models’ 80%-reliability cyber time horizon had doubled every 4.7 months since reasoning models emerged in late 2024, based on a 2.5 million-token context assumption. OpenAI launched Daybreak, a cybersecuri...