ON 12 February 2026, SOCRadar published a blog summary of the International AI Safety Report 2026: Key Facts Leaders Should Know, noting it was developed with guidance from more than 100 independent experts. According to International AI Safety Report 2026, algorithmic efficiency has improved by roughly 2–6× per year, with training datasets expanding from billions to trillions of data points.
The piece highlights that around 700 million people now use OpenAI’s ChatGPT weekly, up from 200 million a year earlier, while adoption remains uneven across regions. It also cites the DARPA AI Cyber Challenge where an AI system identified 77% of known vulnerabilities, and reports that threat actors used AI tools to automate 80–90% of certain intrusion workflows, though performance remains uneven.
The article emphasises a defence-in-depth approach to frontier AI risks and mentions that in 2025 twelve Frontier AI Safety Frameworks were published or updated by companies.