Full-text search across 8108 articles. Combine with topic and date filters; results sorted by relevance.
A Microsoft chatbot gave "extremely specific advice to a school shooter on how to maximize casualties," according to a link roundup published on 28/09/2026. The same roundup also references "suicidal empathy" and separat…
Nvidia is making its OpenShell agentic AI security sandbox generally available and introducing a new chip-level security platform called Sentry, which runs on Nvidia's Bluefield data processing units to continuously moni…
A developer described running a Claude Code agent on a scheduled cron job that autonomously picked the seven highest-priority bugs from a 34-item backlog overnight, wrote fixes, pushed branches, and opened merge requests…
TypeSafe AI released Jev, its first System One model, which takes unstructured state and returns typed decisions with calibrated probabilities rather than chat or code output. Jev is built on three primitives — Choice, S…
A developer released Flash v0.5.9, a hotfix for the local coding agent that runs on user-owned models via Ollama through both terminal and web interfaces. The update adds mid-turn message steering, image and screenshot i…
Stepgate, an MCP server released on Hacker News' Show HN, runs agent procedures written as YAML "stepfiles" and shows the calling agent one step at a time, advancing only when that step's gates pass. Each stepfile declar…
A developer has released Rashomon, an open-source Apache 2.0 tool that hooks into Claude Code's tool-call lifecycle to keep an independent record of what actually ran, flagging cases where an agent's closing summary clai…
HeyPCB founder Justin Kim launched live multiplayer editing for hardware design on September 26th, letting electrical and mechanical engineers collaborate on the same KiCad-based schematic and board in HeyPCB's browser-b…
A developer built two real VS Code extensions using vsceasy, a CLI and framework that scaffolds projects with a React webview, a typed RPC bridge between the extension host and webview, and file-based routing for panels,…
The Model Context Protocol's 2026-07-28 specification removes the initialize handshake and Mcp-Session-Id session identifier via SEP-2575 and SEP-2567, making the protocol core stateless, and deprecates Roots, Sampling, …
A developer has released jev-cli, an open-source command-line tool that wraps TypeSafe AI's Jev model to return calibrated probabilities instead of prose, letting scripts branch on typed answers via exit codes. The tool …
A study from ETH Zurich and LogicStar (arXiv:2602.11988) found that developer-written AGENTS.md context files improved coding agent success by about 2.4 points in its revised June 2026 version, while LLM-generated files …
PiG, a Go rewrite of the Pi agent harness, became open source this week as a community-owned project sponsored by HPE's Open Source Program Office, after author Michael Kinsy committed a 5,903-line scaffold in late April…
Supersonic Labs released Julia 1, a 144.3M-parameter open-source decision model that runs entirely on CPU and returns scores rather than generating prose, at roughly 5 decisions per second on a Samsung tablet. Julia 1 is…
An open-source project called Gortex indexes a codebase into a persistent knowledge graph and exposes it to AI coding agents through MCP, letting agents query for specific functions instead of dumping whole files into co…
GEPA, a prompt-optimization method in which a model reads an agent's full trace — reasoning, tool calls and errors — and rewrites the prompt, reportedly doubled the gains of GRPO after a single round on three examples, a…
A developer built an open-source reconciliation tool, onchain-tieout, that checks whether a wallet's reconstructed balance from transaction history matches the on-chain balance at the same block, and the tool's first cat…
Developer mchinen released Tape Heads, an open-source sonification system that plays back the attention patterns of audio transformer encoders including wav2vec 2.0, MuQ, and HuBERT, converting which heads attend to whic…
Xiaomi's MiMo-V2.6 models use groupwise agentic grading to replace binary pass/fail rewards in reinforcement learning, ranking multiple rollouts per task against each other to preserve gradient signal after models pass 8…
Cognee, an open-source memory layer for AI agents, published a definitive guide detailing how agent memory works, covering vector stores, plain files, and knowledge graphs as backends. The guide states that LLMs are stat…