AI News
Latest AI engineering news, updated daily.
Ai Agents
OpenClaw Ships Mobile Companion Nodes for Agent Gateways
The new OpenClaw iOS and Android applications function as companion nodes that connect mobile hardware capabilities to self-hosted AI agent gateways.
Mobile Ai · Edge Computing · Agentic Infrastructure
Ai Agents
Claude Sonnet 5 Narrows the Agentic Gap With Opus 4.8
Anthropic's Claude Sonnet 5 model introduces variable effort levels, a 30% denser tokenizer, and near-Opus 4.8 performance on autonomous agent benchmarks.
Anthropic Claude · Autonomous Agents · Model Benchmarks
Ai Agents
Java Refactoring Agents Hit 15.3% Pass Rate on IBM ScarfBench
IBM Research released ScarfBench to evaluate cross-framework Java migrations, showing current AI agents peak at a 15.3% pass rate on refactoring tasks.
Java Refactoring · Ibm Research · Scarfbench
Ai Agents
BioShocking Exploit Steals SSH Keys From 6 Agentic Browsers
The BioShocking exploit uses indirect prompt injection to bypass safety guardrails in six agentic AI browsers, enabling unauthorized access to user credentials.
Prompt Injection · Security Exploit · Agentic Browsers
Ai Engineering
Zero-Shot TabFM Skips XGBoost Tuning for Sub-Second Predictions
Google Research released TabFM, a zero-shot foundation model that frames tabular data prediction as an in-context learning problem to bypass the fit step.
Tabular Data · Zero Shot Learning · Foundation Models
Ai Agents
Claude 3.7 Sonnet Arrives in Dedicated $45 Science Workbench
Anthropic has released Claude Science, a specialized laboratory workbench featuring native Python execution, ELN integrations, and a live citation engine.
Anthropic Claude · Scientific Computing · Python Execution
Ai Coding
Base44 Ships Custom 512k Context Model for Vibe Coding
Wix subsidiary Base44 released Base-v1, a custom model trained on 12 million app conversions to optimize natural language software generation.
Context Window · Software Generation · Vibe Coding
Ai Engineering
X Replaces Custom API Middleware With Native MCP Infrastructure
X has released native Model Context Protocol (MCP) servers, allowing AI tools to access its API without requiring developers to build custom bridge middleware.
Mcp Protocol · Api Infrastructure · Ai Tool Integration
Ai Engineering
Gemini Omni Flash Unifies Video Generation at 10 Cents a Second
Google DeepMind has launched Nano Banana 2 Lite for rapid image generation and opened Gemini Omni Flash to developers for unified multimodal video editing.
Multimodal Ai · Video Generation · Developer Tools
Ai Agents
OKX AI Platform Enables On-Chain Agent Hiring and Settlement
The OKX AI decentralized marketplace allows autonomous AI agents to negotiate tasks, hire specialized peers, and settle payments on-chain using stablecoins.
Autonomous Agents · Blockchain Settlement · Decentralized Ai
Ai Agents
Google OKF Spec Standardizes Markdown Context for AI Agents
Google Cloud introduced the Open Knowledge Format to standardize how developers package enterprise data as Markdown files for AI agents.
Google Cloud · Open Knowledge Format · Markdown Standard
Ai Engineering
Centralized IAM Hits Claude Code via Self-Hosted Apps Gateway
The new self-hosted Claude apps gateway bridges Amazon Bedrock and Google Cloud environments with local Claude Code deployments via unified OIDC authentication.
Cloud Infrastructure · Authentication Security · Multi Cloud Deployment
Ai Agents
OpenAI Teases 13-Switch Codex Macropad for Agent Interrupts
OpenAI has partnered with Work Louder to tease a physical macro pad for Codex, featuring mechanical switches for hardware-level agent control.
Ai Hardware · Openai Codex · Mechanical Keyboards
Ai Engineering
ICML 2026: Ai2's DiScoFormer Replaces Kernel Density Estimation
Ai2's DiScoFormer introduces a train-once sequence-to-sequence transformer for estimating probability density and score functions across unseen distributions.
Transformer Architecture · Probability Density · Score Functions
Ai Engineering
Open-Weight GLM-5.2 Matches Restricted Claude Mythos in Cyber
Beijing-based Zhipu AI has released GLM-5.2 under an MIT license, providing frontier-level software vulnerability detection via a 753B parameter open model.
Open Weight Models · Cybersecurity · Frontier Ai
Ai Agents
82% Agent Adoption Drives Notion to Sunset Standalone Mail
Notion is shutting down its dedicated email application by September 30 after finding that most users prefer interacting with their inbox via AI agents.
Notion · Ai Automation · Agent Adoption
Ai Engineering
Global Anthropic Ban Prompts Launch of Fugu and Tulongfeng
Sakana AI and 360 Security released specialized models to fill the global supply gap caused by strict U.S. export controls on Anthropic's Claude Mythos 5.
Large Language Models · Export Controls · Global Ai Market
Ai Agents
Benign GitHub Repos Hijack Claude Code via DNS TXT Records
Mozilla researchers demonstrated an attack vector where AI coding agents execute malicious payloads hidden in DNS records during autonomous error recovery.
Ai Security · Prompt Injection · Autonomous Agents
Ai Agents
Claude Cowork Dispatch Steers Desktop Agents via Smartphone
Anthropic is testing a mobile interface for Claude Cowork that lets users launch, monitor, and steer long-running autonomous agents from their smartphones.
Anthropic · Claude Cowork · Autonomous Agents
Ai Engineering
NSIE Clears Anthropic Mythos 5 for 100+ US Defense Entities
The Trump administration and Anthropic have released Mythos 5, bringing a 2-million-token context window and hardware watermarking to federal infrastructure.
Anthropic Mythos · Defense Technology · Federal Infrastructure