OpenAI Introduces Operator: Browser-Controlling AI Agent
AI agent can interact with web browsers directly, navigating websites and completing multi-step tasks without human intervention.
Google DeepMind Demonstrates Self-Improving Code Agent
Agent iteratively improves its own code through execution feedback, achieving 15-40% performance improvements over human-written baselines.
EU AI Act Establishes Risk-Based Framework for Agent Systems
European Union finalizes implementation guidelines addressing autonomous agent systems with mandatory human oversight for high-risk applications.
Anthropic Publishes Multi-Agent Coordination Safety Research
Research examines risks specific to multi-agent systems, focusing on emergent behaviors and unintended coordination strategies.
Also Tracking
Up to $100k in cloud credits for eligible startups building AI agents.
Free tier + credits for founders building on Azure AI services.
New enterprise accounts receive credits for testing agent workflows.
Free API access for researchers studying agent safety and alignment.
New testing suite evaluates agent performance across real-world task completion scenarios.
Multi-agent system screens molecular compounds and designs synthesis pathways autonomously.
Research on how to specify model behavior for safer and more reliable AI agents.
Benchmark for measuring real-world assistant capabilities across 466 curated questions.