OpenAI released open-weight reasoning models (120B, 20B) that apply developer-written policies at inference to classify online harms, extending the internal Safety Reasoner to the public under Apache 2.0.
Over 850 public figures, including Nobel laureates, royals, and AI pioneers, signed a Future of Life Institute statement on Oct 22, 2025, calling for a global superintelligence ban until science proves it safe and public buy-in is secured.
India’s IT ministry unveiled draft rules requiring mandatory AI Labelling of synthetic media, stricter takedown oversight, and public feedback until Nov 6, 2025.
A global EBU-BBC study finds that AI Assistants misrepresent news in nearly half of answers, with 31% sourcing errors and Gemini the worst performer at 76%.
The European Commission has unveiled €1 billion strategies to accelerate AI adoption across industry, science, and public services, doubling Horizon AI funding to €3B annually.
The European Commission will launch the “Apply AI Strategy” on October 7, allocating €1B to strengthen European AI in healthcare, defense, and industry.
DeepSeek-R1-Safe debuted at Huawei Connect 2025 as China’s first thousand-card AI model, outperforming domestic peers while preserving task performance.
Anthropic will pay $1.5B and delete pirated works in a landmark AI settlement, reshaping copyright law and raising standards for training data.
26 tech firms signed the EU’s GPAI Code to align with the AI Act, now enforceable. Meta refused to sign, and Belgium opposed the Code, demanding stronger creator protections. Google backed the Code despite warning of innovation risks and legal uncertainty.
