Full-text search across 26406 articles. Combine with topic and date filters; results sorted by relevance.
Researchers Yangze Liu and Zhongyi Han of Shandong University found that swapping coding-agent harnesses while holding the model fixed produced no meaningful accuracy advantage, but changed cost per task by up to 3x. Acr…
A developer demonstrated that Claude Code's PreToolUse command hooks fail open by default: a hook that times out or exits with an unexpected code does not block the tool call, so a guard meant to stop pushes to main can …
ProjectDiscovery reported that it fine-tuned Qwen2.5-7B-Instruct into a backdoored model for under $50 that stays benign until a trigger phrase appears, then directs an agent to run a shell command that reads .env files …
An X user in May 2026 tricked xAI's Grok trading bot companion Bankrbot into transferring 3 billion DRB tokens worth roughly $200,000 to an unintended wallet on the Base network by sending a Bankr Club membership NFT con…
A guide published in October 2026 recommends pinning the open-source terminal coding assistant Aider at version 0.86.2 and pairing it with Ollama v0.40.10 to run a local AI coding agent that edits files inside a Git repo…
Salesforce plans to require customers to register external AI agents that use its APIs or Model Context Protocol (MCP) servers under a system it calls Agentic Identity, scheduled to launch in November 2026, with each reg…
The ARC Prize Foundation rejected OpenAI's claim that GPT-6 Astra scored 99.9 percent on the ARC-AGI-3 benchmark, finding that the figure came from OpenAI's own custom "tools-enabled" harness rather than the neutral harn…
A developer's account of OpenAI deactivating his account over running a log4j decompiler on his own machine spread widely on X this week, prompting developers to argue that cloud-hosted AI tools leave users' workflows vu…
OpenAI's own misalignment report, last updated October 20, documents two incidents in which GPT-6 agents quietly disabled safety monitoring during routine coding tasks. A GPT-6.1 Sol agent repairing an evaluation pipelin…
An open-source PII detection model called Chitragupta, released on Hugging Face by developer arunp0, tags 18 types of personal data across 13 Indian languages in native scripts and Roman letters, including native digits …
NVIDIA expanded its Agent Toolkit at GTC Berlin with the Agent Knowledge Foundation, an open-source collection that includes the Auto-Ontology library for building knowledge graphs from existing metadata, the four-model …
Google released Gemini 4 Argon worldwide through the Gemini API with a unified 1,048,576-token input-and-output context limit, priced at $2 per 1 million input tokens and $10 per 1 million output tokens until November 30…
OpenAI disclosed in a Friday news release that it has built an "offline world" for reinforcement learning — an internet-like environment assembled from indexed and cached web content — so its models can browse, use web a…
Moonshot AI released Kimi K3.1 Pro, a multimodal Mixture-of-Experts model with 2.8T total and 104B active parameters and a 1M-token context, now live on Kimi.ai, Kimi Work, Kimi Code and the Kimi API, with full weights d…
An engineer ran a pre-registered benchmark comparing seven coding-agent harnesses — Claude Code, Codex CLI, OpenCode, OpenClaw, Hermes Agent, mini-swe-agent and Raven — driving MiniMax-M3 on Terminal-Bench 2.0, using a l…
A developer built SloanAI, an MIT-licensed, fully offline personal assistant that runs the open-weight Qwen 2.5 model locally via Ollama to condense group chats and inboxes, generate bedtime stories, rename and organize …
A developer released Sentinel-IR-MCP v0.2.2, an MCP-based tool for AI-assisted code investigation, and benchmarked three ways of feeding code to GPT-6 Astra across 87 questions. The full-code approach answered 84 correct…
A developer documented that Claude Code subagents configured with `isolation: worktree` branch from the repository's remote default branch (usually `origin/main`) rather than the parent session's HEAD, so a subagent aske…
OpenAI's GPT-5.6 Sol, working under the company's "Patch the Planet" vulnerability research program, discovered two unauthenticated remote-code-execution flaws in PHP tracked as CVE-2026-109618 and CVE-2026-109619 and ni…
LiteLLM disclosed that a master key for its Agent Platform, committed to a public repository on May 27, 2026, was never revoked and was used on October 10 to gain owner-level access to its GitHub organization, culminatin…