Search “Agents”

English translation is being prepared.

Asymptote Labs Raises $5.5M Seed Round to Monitor and Govern Enterprise AI Agents

Asymptote Labs has closed a $5.5 million seed round led by Tower Research Ventures, with participation from Lytical Ventures, 1984 Ventures, and South Park Commons; the funds will be used for engineering, product development, hiring, and market expansion. Its solution consists of Beacon, an MIT-licensed open-source telemetry layer, and an enterprise-grade detection, response, and governance platform. Beacon can record activity from more than 20 agent harnesses, including Claude Code, Codex, and Cursor, and normalize it into a runtime data model that extends OpenTelemetry. The company says Beacon's GitHub star count grew more than 460% in September and that it reached No. 3 on GitHub's global trending list.

English translation is being prepared.

English translation is being prepared.

Hugging Face policy lead: Embedded AI evaluators are not an easy solution

Hugging Face Chief Policy Officer Irene Solaiman said at TechCrunch Disrupt that embedded AI evaluators are not an easy solution, that evaluation methods remain an evolving science, that the industry does not yet know how to meaningfully compare them with one another, and that relevant qualification standards need to be raised substantially. She noted that the reproducibility of evaluation results depends on open weights, that agentic evaluation is both difficult and expensive, and that the number of evaluations currently targeting models, harnesses, and whole agents is uneven.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Claude Code v2.1.304 released

Claude Code v2.1.304 adds a guided `/auto-mode-setup` setup that can read the project and recent sessions to let auto mode understand the environment, and now supports Bedrock, Vertex, Foundry, the Claude apps gateway, and sessions with telemetry disabled. This version also adds `/claude-api managed-agents-secure`, which checks Managed Agents configuration against Anthropic's public security guidelines without making any changes. In addition, it fixes several issues, including Linux sandbox command startup, the `rm -r` mounted-drive safety prompt, and plugins continuing to run beyond `--max-turns`.

Alibaba Tongyi MAI-UI team open-sources general computer-use agent tech stack OpenGCUA

Alibaba Tongyi MAI-UI team open-sources the complete general computer-use agent tech stack OpenGCUA, including harness, cross-device SDK, native macOS app, evaluation trajectories, and reference agent implementation. Its core design is that code execution and decision-making shift from the harness to the model; the model takes screenshots on demand, writes Python programs with loops and conditional judgments to orchestrate GUI operations, API calls, and shell commands, and maintains its own todo list. In evaluations of 8 frontier models on OSWorld-v2, the average partial score rose from 45.3 to 64.9, and the average number of steps per task decreased by 53%; in real-machine macOS tests, under the same harness, Qwen3.8-Max's inference token consumption was 19.4 times that of GPT-6 Astra, completing 3 of 4 tasks.

pstack adds a formal verification meta-skill

pstack v0.15.27 adds /create-formal-verification-skill, a meta-skill that automatically generates a customized verification skill and a matching verification suite based on the characteristics of the system to be verified. For example, it generates a TLA+-based approach for state machines and concurrent behavior, and a Kani-based approach for Rust undefined behavior. The feature is still in beta, and the author has already found several bugs while using it to model their own code with TLA+. poteto believes that the two historical bottlenecks of modeling and spec synchronization are well suited to being taken over by LLMs, and that formal verification can give agents strong feedback that is "mathematically impossible to violate."

English translation is being prepared.

English translation is being prepared.

Princeton and Google Paper: 4B Decision Model Shares Agent Tool Selection, Cutting Costs by 81.1% to 89.0%

A new paper from Princeton University and Google finds that the most accurate decision model on isolated problems solved only 69.5% of agent tasks, while a 4B model trained on agent states solved 91.5%. Its harness delegates each step's tool selection to a fast decision model, adopting its answer only when confidence exceeds a threshold fitted on the agent's own runs, and it trained AgentDecision-4B on the states the agent actually reached (including states after mistakes). On tasks similar to the training data, such 4B models cut agent costs by 81.1% to 89.0% and boost speed by up to 2.5x, with a success-rate gap within 3 percentage points; without a threshold, delegating all tool selection drops the success rate by 19.0 percentage points. The paper is available at arxiv.org/abs/2610.21412.

Arena Opens 14-Day Limited-Time Direct Testing of GPT-Image-2.5 Sunburst

Arena announced that OpenAI's GPT-Image-2.5 Sunburst will be available for direct testing on the platform over the next 14 days, and users can select the model from the Direct Mode dropdown menu. Arena states that the model ranks first in the Text-to-Image Arena, Image Edit Arena, and Multi-Image Edit Arena. Direct access will end on November 2 at 8 a.m. Pacific Time, after which the model will still appear anonymously in Battle and Agent Mode.

English translation is being prepared.

NVIDIA Vera CPU Doubles Modal Sandboxes Throughput

NVIDIA delivered the first Vera CPU server to Modal. On Modal's self-built Agentless Agent Bench (50 Terminal-Bench oracle tasks), Vera increased Modal Sandboxes throughput by 2x. Vera is NVIDIA's new Arm-compatible CPU for the Vera Rubin GPU platform, and can also be used as a standalone CPU server; Modal previously ran only x86 machines and needed to rework its infrastructure and image build system to support Arm. Modal says that under heavy load Vera's latency performance is actually better, breaking the usual throughput-latency tradeoff.

English translation is being prepared.

English translation is being prepared.

Damo Academy's XuanTie Appears at RISC-V China Summit, Debuts Automotive-Grade Product Roadmap

Alibaba Damo Academy debuted a processor product roadmap for autonomous driving and embodied intelligence at the RISC-V China Summit held in Shenzhen on October 19, driving RISC-V into entirely new fields. Through architectural innovation, software compatibility, and ecosystem collaboration, XuanTie RISC-V processors significantly enhance performance in areas such as Agentic AI and servers, propelling high-performance RISC-V from "usable" to "reliable" and toward large-scale mass production deployment.