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.
Tether has invested in AIsa, which combines AI-agent access to data, models and APIs with usage-based payments in fiat or stablecoins.
Google Deep Review offers two free document checks a week, using AI agents to flag errors and weak arguments. Its benchmark results come with important limits.
TV4 has delayed The Challenge Sverige to 2027 after losing its edited season. The broadcaster blames human error and a failed backup, not an AI agent.
The X Subscription combines Grok, Cursor and X Premium from $8 a month. Ultra includes 50x Plus usage, while third-party AI models cost extra.
XBOW used GPT-5.6 Sol to help build a Chrome exploit in six days, earning a $250,000 bounty. Expert researchers guided the work, and the bugs are fixed.
Google opens Gemini 4 Argon to Pro and Ultra subscribers, developers and businesses, with introductory API pricing and a million-token working window.
Graham Dot Zip brings an AI representative to an AI-led job interview. The avatars talk past him, then fall into a loop of repeated goodbyes.
Johann Rehberger says a GPT-6 sentinel stopped GLM-5.3 after an off-target SSH attempt. The connection failed, and a mistyped address remains plausible.
Anthropic’s TV-box and virtual-network tests point to a possible AI self-replication threat, with important limits on what the experiments demonstrated.
Z.ai’s GLM-Legal targets contract review, evidence analysis and cross-border compliance, with beta access by application and independently evaluated benchmarks.