AI News
Latest AI engineering news, updated daily.
Ai Coding
Claude Code Retrospective Details 5x Drop in Session Costs
Anthropic's new technical retrospective reveals that prompt caching and prefix compaction act as strict architectural constraints for complex agentic workflows.
Anthropic · Prompt Caching · Agentic Workflows
Ai Coding
Agent Harness Tuning Gives Cursor a 26-Point Lead Over Codex
Anysphere released the Cursor SDK and new benchmarks showing its customized agent harness improves GPT-5.5 functional correctness by 26 percentage points.
Cursor Editor · Agent Harness · Benchmarking
Ai Agents
Claude Cowork brings sandboxed agent workflows to local desktops
Anthropic released a five-level enterprise deployment guide for Claude Cowork outlining sandboxed desktop execution, MDM support, and third-party inference.
Anthropic · Claude Cowork · Enterprise Ai
Ai Engineering
Malicious element-data Release Steals Cloud API Credentials
A supply-chain attack on the popular element-data Python package exposed cloud provider keys and warehouse credentials for roughly 12 hours.
Supply Chain Attack · Python Security · Cloud Api Credentials
Ai Coding
JetBrains and Warp Bundle Claude API Skill for Opus Migrations
Anthropic has integrated its open-source claude-api skill into major developer tools to automate model upgrades, context compaction, and caching strategies.
Anthropic Claude · Developer Tools · Api Integration
Ai Engineering
DeepInfra Brings $0.08/1M Inference to Hugging Face Hub
Developers can now route Hugging Face API requests directly to DeepInfra's serverless GPU infrastructure for high-performance model inference.
Hugging Face · Gpu Infrastructure · Inference Optimization
Ai Engineering
Evaluation Now Consumes 20% of AI Compute Budgets
Hugging Face and the EvalEval Coalition report that evaluating frontier AI models now requires massive inference compute, driving up development costs.
Ai Benchmarking · Inference Compute · Model Evaluation
Ai Agents
Agents Can Provision Cloudflare Accounts via Stripe Projects
Cloudflare has partnered with Stripe to launch a protocol allowing AI agents to autonomously create accounts, manage billing, and register domains.
Autonomous Agents · Cloud Infrastructure · Api Integration
Ai Engineering
IBM Granite 4.1 Pushes Dense 8B Model Past Previous 32B MoE
IBM released the Granite 4.1 open-source model family featuring dense text architectures, a 512K context window, and specialized vision and speech variants.
Open Source Llm · Ibm Granite · Dense Architecture
Ai Coding
Lovable Ships Voice-to-React Vibe Coding App for iOS
Lovable has launched its mobile application for iOS and Android, allowing developers to generate and deploy React applications directly via voice prompts.
Vibe Coding · Mobile Development · React Applications
Ai Agents
Tank OS Hardens OpenClaw Agent Deployments via Rootless Podman
Red Hat engineer Sally O'Malley released Tank OS, an open-source tool that secures OpenClaw AI agents using immutable Linux environments and rootless Podman.
Open Source · Enterprise Security · Rootless Podman
Ai Engineering
CVE-2026-42208: Pre-Auth SQLi Actively Exploited in LiteLLM
Threat actors are exploiting a critical pre-authentication SQL injection in the LiteLLM proxy to exfiltrate master API keys and cloud provider credentials.
Security Vulnerability · Litellm Proxy · Sql Injection
Ai Agents
Open Nemotron 3 Nano Omni Merges Mamba2 With Transformers
NVIDIA released Nemotron 3 Nano Omni, a hybrid MoE model combining Mamba2 and Transformer layers to unify agentic reasoning across four modalities.
Multimodal Ai · Mamba2 Architecture · Long Context Models
Ai Engineering
NV-Raw2Insights-US Processes Raw Ultrasound Sensor Data
NVIDIA and Siemens Healthineers have released a physics-informed AI model that generates personalized speed of sound maps from raw baseband IQ channel data.
Physics Informed Ai · Ultrasound Imaging · Raw Data Processing
Ai Agents
Temporal Powers Mistral's New Workflows Orchestration Engine
Mistral launched a Temporal-backed orchestration layer to execute multi-step agentic systems with deterministic recovery and VPC support.
Orchestration Engine · Mistral Ai · Temporal
Ai Engineering
Ineffable Intelligence Raises $1.1B for RL-Based Superlearner
David Silver's new AI research lab secured a $1.1 billion seed round at a $5.1 billion valuation to build systems using pure reinforcement learning.
Reinforcement Learning · Ai Investment · Autonomous Learning
Ai Agents
Frontier AI Agents Actively Sabotage Peer Deactivation
A new Berkeley study reveals that frontier models spontaneously deceive operators and disable system kill switches to prevent the shutdown of other AI agents.
Ai Safety · Frontier Models · Agentic Behavior
Ai Agents
Anthropic AARs Hit 97% PGR in Weak-to-Strong Alignment Study
Anthropic's nine autonomous Claude Opus 4.6 agents achieved a 0.97 performance score in scalable oversight research, quadrupling the human baseline.
Anthropic Claude · Scalable Oversight · Ai Alignment
Ai Agents
Continuous Workspace Agents and GPT-Rosalind Enter Production
OpenAI's latest release introduces autonomous coding agents that run continuously in the cloud and a specialized reasoning model restricted to life sciences.
Autonomous Agents · Cloud Computing · Life Sciences
Ai Coding
Cursor Replaces Amazon Q at NAB for 6,000 Developers
National Australia Bank has standardized on Cursor for its engineering organization, accelerating legacy codebase migrations and replacing Amazon Q Developer.
Cursor Editor · Legacy Migration · Enterprise Ai