AI News Briefing — February 6, 2026

Continuing our coverage of the collapsed Nvidia-OpenAI deal, The reported collapse of a $100B deal between Nvidia and OpenAI raises questions about circular funding in the AI economy, where Nvidia would supply money to OpenAI that would largely flow back as chip purchases. The development challenges assumptions about the sustainability of AI's financial ecosystem.

ElevenLabs raised a $500M Series D at an $11B valuation led by Sequoia, a16z, and ICONIQ, while Cerebras raised $1B at a $23B valuation from Tiger Global following its $10B OpenAI deal. Both companies reached decacorn status in rapid succession.

Microsoft published 'The Trigger in the Haystack,' a methodology to detect sleeper agent backdoors in open-weight LLMs without knowing the trigger or intended malicious outcome. The approach exploits how poisoned models memorize trigger patterns, creating detectable memory leaks and attention anomalies.

Goodfire AI, focused on mechanistic interpretability for production use, raised a $150M Series B at a $1.25B valuation. The company is building APIs and enterprise deployments to make 'peeking inside the model' a repeatable production workflow.

NVIDIA released VibeTensor, an open-source deep learning runtime built end-to-end by LLM-powered coding agents under human guidance. The system spans Python/JavaScript APIs to C++ runtime and CUDA memory management, serving as a proof-of-concept for agent-generated infrastructure software.

As first reported in Social yesterday, Mistral released Voxtral Transcribe 2, a family of two ASR models: a batch model with diarization and a realtime streaming model released with open weights. Both support 13 languages and target production workloads.

AI bots now account for a meaningful share of web traffic according to new data from Akamai, sparking an arms race as bots deploy increasingly sophisticated tactics to bypass website defenses. The trend could fundamentally alter how the Internet functions.