AI Tools news and analysis on Web Pulse: 38029 curated articles tracking the latest AI Tools developments, tools, and research, updated continuously from vetted sources.
A September 15, 2026 arXiv paper demonstrates that poisoned benchmarks can induce self-modifying AI coding agents to write vulnerable code on clean, held-out tasks, with Hyperagents powered by Sonnet 4.5 self-evolving in…
A scan found 230 of 243 unauthenticated LocalAI instances vulnerable to exploitation, a flaw recently exploited in a large-scale campaign that compromised high-value targets including infrastructure linked to the Thai mi…
A new arXiv paper (2609.19476v1) reports a latent space Bayesian optimization method that achieves at least a 100x speedup over state-of-the-art baselines while matching or improving performance on molecular and image ge…
A new arXiv paper (2609.19526v1) introduces Recursive Self Improvement via Fast Tree-search (SIFT), a sample-efficient framework that lets coding agents recursively modify their own implementations under strict budget co…
A cooperative framework called UnifiedPlayers, comprising a Planning Player, an Execution Player, and an Evaluation Player, outperformed the strongest prior baseline by at least 3.5% on mathematical reasoning and 3.9% on…
A new arXiv paper (2609.20152v1) introduces MTVA-Bench, a benchmark that evaluates the language model inside cascaded voice agents under the same conditions it faces in production, covering 49 agents, 490 reviewed scenar…
A decomposed confidence layer combining perception, layout, and validation channels with conformal risk control raised AUROC for financial-document key-value extraction from 0.54-0.74 with native Vision Language Model ve…
Graham Dumpleton published a follow-up post demonstrating how to build a complete JupyterLab workshop from an empty directory using the jupyterlab-workshop package's `jupyter workshop init` command, which scaffolds a man…
Mozilla and Mistral announced a partnership bringing the 119-billion-parameter Mistral Small 4 model to Firefox Smart Window, replacing OpenAI's gpt-oss-120B in the feature's "Personalized" category. Firefox 155 opened S…
A developer has released @coder12-z/llm-shield, an MIT-licensed, zero-dependency Node.js library that wraps LLM API calls to handle intermittent failures such as transient 400 "could not parse JSON body" errors, truncate…
A developer has published a from-scratch Python walkthrough for building a retrieval-augmented generation (RAG) system, covering chunking, embeddings, a minimal vector store, hybrid retrieval, re-ranking, and cited gener…
A developer released a free, MIT-licensed GitHub Action that scans pull request diffs for malicious code using regex and semgrep rather than an LLM, targeting command execution, credential access, exfiltration, auto-run …
PrismML released Bonsai 2 27B on September 17, 2026, a ternary-quantized version of Alibaba's open Qwen3.8-27B that retains 98.2% of the full-precision model's benchmark average while shrinking from 54GB to 5.9GB. The mo…
A forum thread is crowdsourcing real-world local LLM deployment data across five VRAM tiers — ≤16GB, 24–32GB, 48–64GB, 96–128GB, and 196–256GB+ — asking users to report model, quantization (AutoRound W4A16, GGUF Q4_K_M/I…
Cognition launched Code Scans, a Devin feature powered by its Agentic MapReduce architecture that turns engineering goals into codebase investigations and pull requests. In a test on the Dioxus repository, Code Scans cut…
A developer released pod v0.3.2, an Apache-2.0 tool that gives AI coding agents a hash-chained audit trail for incident forensics. The tool records each tool call with an argument hash, decision, outcome, and the policy …
Lodestar released version 0.1.121, a macOS workspace for Claude Code and Codex that lets users keep a shared thread across multiple AI agents. The app maps connections across a session and its children, tracks usage limi…
A developer has released an open-source project called Resume the Scene that preserves a project's verified execution point independently of chat history or model vendor, arguing that AI continuity failures stem from mis…
Apple's WebKit blog detailed Safari 27.0, which adds a Safari MCP server that gives coding agents such as Claude Code and Codex control over the browser window to inspect the DOM, network requests, screenshots, and conso…
A developer's guide argues that AI-generated video ads often sell a mood while failing to make the product recognizable, and proposes a three-second, audio-off test to catch the problem before polishing the wrong shots. …