TAU-bench (Tool-Agent-User benchmark) is an evaluation framework designed to test how effectively AI agents can interact with users and external tools in realistic, multi-turn scenarios. Developed by researchers from Anthropic, it measures an agent’s ability to complete structured tasks—like booking flights or managing online orders—while following rules, handling APIs, and reasoning across multiple steps. A key innovation in TAU-bench is its pass@k metric, which evaluates not just whether an agent succeeds once, but how consistently it can repeat correct behavior across several trials. The benchmark highlights that even advanced AI models often struggle with reliability, consistency, and proper tool use over extended interactions. TAU-bench has quickly become a standard for assessing the real-world robustness of autonomous and conversational AI systems.
AMD Helion offers CPU-only and integrated CPU/GPU racks for AI agents, with up to 32,768 Zen 6 cores and November 2026 availability planned.
Microsoft CEO Satya Nadella wants humans to be able to stop advanced AI mid-task, with independent controls, audit trails and limits on agent access.
Odyssey-3 generates worlds that respond to actions, leads a published physics benchmark and powers robot demonstrations. Flash is available to try in a browser.
Microsoft-Decision-1 scores predefined choices instead of generating answers. It is available on Foundry and OpenRouter at $0.042 per million input tokens.
Anthropic details unintended Claude agent actions on government websites. Philadelphia police says a fabricated homicide tip was blocked as spam.
AI companies are rehearsing potential crisis responses, Axios reports. OpenAI says the scenarios are not inevitable, while recent incidents test safeguards.
Google Cloud’s Gemini agent can create subagents and use Claude alongside Gemini. Background delegation and third-party models remain in early access.
Manus raises more than $500 million in its first funding round since separating from Meta, with Boyu Capital and IDG Capital leading the financing.
Claude Haiku 5.5 cuts costs for everyday AI tasks, with API rates up to 20x below Sonnet and Anthropic-reported benchmark gains over GPT-6 Luna.
BlackRock sees stablecoins as a potential payment method for AI agents buying data and compute, while the Bitcoin savings thesis rests on simulations.