BREAKINGOpenAI closes $40B round at $340B valuation — largest private tech raise ever·MODELSAnthropic ships Claude Opus 4 with extended thinking and agentic capabilities·FUNDINGxAI raises $6B Series C led by Andreessen Horowitz for Grok infrastructure·REGULATIONEU AI Act enters full enforcement — high-risk systems must comply now·AGENTSGoogle DeepMind open-sources Gemini Agent Framework for autonomous task completion·RESEARCHStanford HAI: Enterprise AI adoption hits 78% globally, GenAI in production at 45%·WARNINGUS Senate passes AI Transparency Act — content labeling required at scale·PRODUCTMeta releases Llama 4 Maverick open-weight model rivaling proprietary alternatives·MODELSDeepSeek V3 scores within 2% of GPT-4o on MMLU at 1/10th the inference cost·FUNDINGMistral AI raises €600M Series B at €6B valuation for European AI sovereignty·BREAKINGOpenAI closes $40B round at $340B valuation — largest private tech raise ever·MODELSAnthropic ships Claude Opus 4 with extended thinking and agentic capabilities·FUNDINGxAI raises $6B Series C led by Andreessen Horowitz for Grok infrastructure·REGULATIONEU AI Act enters full enforcement — high-risk systems must comply now·AGENTSGoogle DeepMind open-sources Gemini Agent Framework for autonomous task completion·RESEARCHStanford HAI: Enterprise AI adoption hits 78% globally, GenAI in production at 45%·WARNINGUS Senate passes AI Transparency Act — content labeling required at scale·PRODUCTMeta releases Llama 4 Maverick open-weight model rivaling proprietary alternatives·MODELSDeepSeek V3 scores within 2% of GPT-4o on MMLU at 1/10th the inference cost·FUNDINGMistral AI raises €600M Series B at €6B valuation for European AI sovereignty·
announcement
Predicting When RL Training Breaks Chain-of-Thought Monitorability
Apr 1, 2026Google DeepMind
Event Summary
Crossposted from the DeepMind Safety Research Medium Blog. Read our full paper about this topic by Max Kaufmann, David Lindner, Roland S. Zimmermann, and Rohin Shah. Overseeing AI agents by reading their intermediate reasoning “scratchpad” is a promising tool for AI safety. This approach, known as C
Related Signals
Anthropic, OpenAI +37 more: 111 model releases in rapid succession
modelsApr 2, 2026
Anthropic, OpenAI +36 more: 107 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +35 more: 107 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +35 more: 107 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +36 more: 111 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +36 more: 117 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +35 more: 119 model releases in rapid successionmodelsApr 2, 2026
Microsoft, Anthropic +35 more: 125 model releases in rapid successionmodelsApr 2, 2026
Anthropic, Microsoft +35 more: 132 model releases in rapid successionmodelsApr 2, 2026
Anthropic, Microsoft +35 more: 137 model releases in rapid successionmodelsApr 2, 2026
Source
Source articles are linked automatically as the intelligence pipeline processes corroborating evidence.