RoboHarm benchmark finds leading AI models unreliable at rejecting dangerous robot commands
RoboHarm testing reported by The Decoder found GPT-6 Astra, Claude Fable 5.1, and MolmoAct2 unreliable at refusing dangerous robot commands.
AI News from September 20, 2026
Latest Artificial Intelligence Updates and Stories
September 20, 2026
Anthropic Reportedly Weighs New Model After Calling for Slower AI Development
Anthropic is reportedly considering a new AI model after urging slower development, potentially intensifying competitive pressure on OpenAI and regulators.
Tencent’s Gander separates conversation from background AI work
Tencent’s Gander separates real-time conversation from background agent work, promising fewer interruptions while exposing tradeoffs in accuracy and multimodal understanding.
Qwen3.8-Omni-Flash Targets Gemini Flash With Lower Multimodal API Costs
Alibaba’s Qwen3.8-Omni-Flash pairs audio-video processing and agent tools with pricing far below Gemini Flash, intensifying multimodal model competition.
Zuckerberg report highlights AI regulation split, but details of OpenAI and Anthropic message remain unclear
A Times of India report says Mark Zuckerberg sees a split over AI regulation, but missing article text leaves his message to OpenAI and Anthropic unclear.
Chinese AI Models Gain Valuation Momentum, but Revenue Trails OpenAI and Anthropic
Chinese AI models are attracting high valuations, but an Invezz report says their revenue remains a fraction of OpenAI and Anthropic’s, exposing a scale gap.
AI Week in Review 26.09.19: No Verifiable Event Details in Available Record
A listed AI Week in Review entry offers no accessible article text, leaving its reported events, claims, and significance unverified for AI industry readers.
A China-US Focus Commentary Calls for Multipolar Cooperation on AI Risks
A China-US Focus commentary links the global AI race to existential risk and argues that cooperation across multiple powers is needed for safer development.
Ads