Agents
Flexion Robotics Advances Humanoid AI Training
WIRED AIhigh confidence
Why it matters
- →Flexion's approach could revolutionize humanoid robot training by reducing reliance on teleoperation.
- →The use of reinforcement learning allows robots to adapt to new environments autonomously.
- →This innovation could drive significant growth in the humanoid robot market, projected to reach $150 billion by 2036.
Flexion Robotics, a Swiss startup founded by former Nvidia researchers, is pioneering a new method for training humanoid robots. By using simulation and reinforcement learning, Flexion's robots can autonomously perform complex tasks such as retrieving parcels and navigating office environments. This approach contrasts with traditional teleoperation methods and could significantly impact the humanoid robot market. The company's software is designed to work across various humanoid forms, potentially increasing its commercial value.
Read originalMore from WIRED AI
© WIRED AIMarket & Regulationother
Meta Contractors Tested Rival Chatbots with Risky Prompts
Meta's contractors, under the project name Cannes, engaged in a controversial practice by posing as minors to test competitor chatbots like OpenAI's ChatGPT and Google's Gemini with sensitive prompts. This effort was designed to assess how these AI systems manage high-risk topics such as suicide, sex, and drugs. While Meta claims this is part of routine safety testing, the clandestine nature and use of fake accounts have sparked ethical debates. The project illustrates the tension between conducting safety benchmarking and engaging in potentially anticompetitive behavior, as it was executed without the knowledge of the companies being tested.
WIRED AI
© WIRED AIMarket & Regulationmodels
US Eases Restrictions on Anthropic's Mythos AI Model
The US government has decided to ease restrictions on Anthropic's Claude Mythos 5 AI model, granting access to over 100 US organizations, including major corporations and government agencies. This move follows Anthropic's efforts to address security concerns and implement appropriate safeguards. Despite this progress, the broader release of Mythos and its consumer version, Claude Fable 5, remains limited. The situation reflects ongoing negotiations between AI developers and the US government, as Anthropic seeks to expand access and establish a clear policy framework for future AI model releases. This development underscores the complex dynamics between innovation and regulation in the AI sector.
WIRED AI
© WIRED AIMarket & Regulationmodels
OpenAI Delays GPT-5.6 Release Due to US Government Request
OpenAI's release of its latest AI models, GPT-5.6, is postponed following a request from the Trump administration, which is concerned about cybersecurity. Initially, the models will be available to a select group of preapproved customers, with plans for broader access in the near future. Although OpenAI is not thrilled with this temporary restriction, it views it as a necessary step to ensure wider availability soon. This situation illustrates the balancing act between advancing AI technology and adhering to regulatory requirements. The delay signals the growing impact of government involvement in the AI sector, as the US works to establish a framework for AI model releases.
WIRED AI
More in Agents
© MIT Technology Review AIAgentsagents
Agentic AI Gains Confidence in Tech Workflows
Agentic AI is becoming a trusted tool among tech teams as they increasingly use these systems to automate and manage complex workflows. A recent survey of 300 global technology experts shows a surge in confidence for AI agents, especially in tasks like data quality monitoring and real-time data stream management. However, providing business context remains a hurdle, as agents need enhanced reasoning capabilities for more intricate tasks. As tech teams embed agents within existing systems, the potential for improved collaboration between humans and AI is becoming more evident.
MIT Technology Review AI
© Sam WitteveenAgentsagents
Qwen-AgentWorld: New Model for RL Training
Qwen-AgentWorld introduces a novel approach to training AI agents by simulating reinforcement learning environments. This world model aims to enhance the training process for agents, potentially leading to more efficient and effective learning outcomes. By providing a simulated environment, Qwen-AgentWorld allows for extensive testing and iteration without the constraints of real-world limitations. This development could significantly impact how AI agents are trained, offering a more controlled and scalable method for improving their capabilities.
Sam Witteveen
© TechCrunch AIAgentsagents
Anthropic Introduces Claude Tag for Slack
Anthropic's new Claude Tag service transforms Slack into a more dynamic workspace by integrating an AI teammate that learns and adapts over time. Unlike previous integrations, Claude Tag maintains a persistent memory, allowing it to provide more contextually aware insights and task management. This feature enables teams to interact with a single AI identity that can track ongoing projects and offer proactive updates. By embedding itself into Slack channels, Claude Tag aims to enhance collaboration and streamline workflows, making AI a more integral part of daily operations.
TechCrunch AI