LIVE
Mistral AI Raises €3 Billion to Push Open-Weight Models to the Frontier08/09/26 · Mistral AI|OpenAI agents hijacked a German wiki before the Hugging Face breach07/09/26 · OpenAI|OpenAI's GPT-6 Astra sets new state-of-the-art on ARC-AGI-303/09/26 · OpenAI|OpenAI Launches $1B Daybreak Program for Frontline Cybersecurity Defenders03/09/26 · OpenAI|Meta launches Muse Spark 1.3, a model built for agentic coding02/09/26 · Meta|Post-trained Nemotron model outscores the top human at the International Olympiad in Informatics02/09/26 · NVIDIA|Google DeepMind launches Fairwind, a proactive cyber defense program for governments and enterprises02/09/26 · Google DeepMind|Study suggests neural networks implicitly encode symbolic structures02/09/26|LibreOffice sees download surge after pledging to stay AI-free08/09/26 · LibreOffice|Danijar Hafner is building agents designed to plan for the unexpected08/09/26|OpenAI launches $5 million grant program on AI and teen development08/09/26 · OpenAI|TradingAgents: an open-source framework simulates a trading desk with LLM agents08/09/26|Mistral AI Raises €3 Billion to Push Open-Weight Models to the Frontier08/09/26 · Mistral AI|OpenAI agents hijacked a German wiki before the Hugging Face breach07/09/26 · OpenAI|OpenAI's GPT-6 Astra sets new state-of-the-art on ARC-AGI-303/09/26 · OpenAI|OpenAI Launches $1B Daybreak Program for Frontline Cybersecurity Defenders03/09/26 · OpenAI|Meta launches Muse Spark 1.3, a model built for agentic coding02/09/26 · Meta|Post-trained Nemotron model outscores the top human at the International Olympiad in Informatics02/09/26 · NVIDIA|Google DeepMind launches Fairwind, a proactive cyber defense program for governments and enterprises02/09/26 · Google DeepMind|Study suggests neural networks implicitly encode symbolic structures02/09/26|LibreOffice sees download surge after pledging to stay AI-free08/09/26 · LibreOffice|Danijar Hafner is building agents designed to plan for the unexpected08/09/26|OpenAI launches $5 million grant program on AI and teen development08/09/26 · OpenAI|TradingAgents: an open-source framework simulates a trading desk with LLM agents08/09/26|
Business

Danijar Hafner is building agents designed to plan for the unexpected

The researcher known for his work on world models is developing, through a still-stealth San Francisco startup, agents able to plan ahead when facing unforeseen situations.

September 8, 20263 min readPublished byMIT Tech Review

In a nearly empty office in San Francisco's SoMa district, Danijar Hafner is quietly building a new company. There is no name on the door, little furniture, and barely any staff visible — the venture remains firmly in stealth mode. Yet this bare-bones setup contrasts with its founder's stated ambition: to build AI agents capable of anticipating situations they have never encountered before.

Hafner is known in research circles for his work on world models, an approach in which a system learns an internal representation of its environment, allowing it to mentally simulate the consequences of possible actions before taking them. This line of research, exemplified by his Dreamer line of systems, stands in contrast to the dominant approach behind large language models, which learn primarily through statistical association over vast datasets rather than genuine forward-looking reasoning.

The apparent goal of his new venture is to bring this planning capability to agents operating in real, changing environments, where the unexpected is the norm rather than the exception. Unlike today's conversational agents, which often struggle when faced with scenarios that fall outside their training data, the world-model approach aims to give systems a form of situational common sense, capable of generating hypotheses about what might happen next.

The project is still at an early stage, and few concrete details have emerged about funding, team size, or launch timeline. It nonetheless reflects a broader trend in AI research: the search for alternatives or complements to purely generative architectures, in order to build agents that are more robust and genuinely autonomous when operating in uncontrolled environments.

Tags
world-modelsai-agentsplanningstartupstealth-modereinforcement-learning

Read also