Full-text search across 70358 articles. Combine with topic and date filters; results sorted by relevance.
A developer outlined three production defenses for hardening tool-enabled AI agents against indirect prompt injection and tool hijacking, framing the risk as a remote code execution and data exfiltration surface rather t…
Google has marked Nano Banana 2 (gemini-3.1-flash-image) for retirement, with an earliest possible shutdown date of 2026-10-29 that the provider notes is estimated and may move. The model, which handles text, image, vide…
Commerce.com, working with Cloudflare, has enabled WebMCP across nearly 100,000 storefronts, letting AI agents shop via structured tools such as search_products and add_to_cart instead of parsing HTML and guessing at con…
Researchers introduced Agent4Sci, a long-horizon testbed for measuring how frontier coding agents perform automated machine learning for science, arguing that improving a scientific model often requires domain insight be…
A new arXiv paper (2610.28811v1) reports that LLM judges used for automated failure attribution in multi-agent systems encode enough internal information to identify the failing step but fail to say it, a gap the authors…
A study posted as arXiv:2610.29162v1 found that all nine frontier and near-frontier AI models tested use more inanimate subjects with animate verbs and more low-frequency n-grams than human writing, and most also produce…
A study posted to arXiv (2610.29328v1) by researchers at the Technion-cs-nlp group found that agreement between two LLMs in multi-turn conversations follows Markovian dynamics when pooled across conversations, even thoug…
An arXiv paper (2610.29052v1) reports that an in-house "System 1" judge built on a frozen backbone with a rank-16 LoRA adapter and a pointer head triaged about 148K agentic coding session records in one forward pass, jud…
A new arXiv paper (2610.28707v1) introduces SimHack, a simulator that predicts when detectable reward hacking will become dominant in an LLM coding RL run before training begins, learning the dynamics from a handful of s…
A framework called DynaWeave that converts static LLM agent training tasks into verifiable dynamic scenarios raised task completion on CostBench by 67.34 percentage points and outperformed Gemini-3.1-Pro-Preview on ToolM…
A new arXiv paper (2610.28714v1) finds that simple baselines come within a median of 2.3 percentage points of the best reported GraphLLM results: Correct & Smooth on TF-IDF features trails each paper's highest reported L…
A Graph Neural Network trained on the Music Notation Graph structure to predict context-dependent glyph offsets cut positioning RMSE by an additional 50 to 85% on stems and beams compared with empirical means, according …
Researchers released KVLearn, a learning-based framework for KV-cache retention in disaggregated LLM serving, reporting end-to-end time-to-first-token reductions of up to 56% versus No-Cache, 38% versus LRU-Pool, and 33%…
A developer built memgraph, a zero-dependency Node.js CLI that stores decisions, facts and their relationships in a single JSON file committed alongside code, so agents and teams can trace why a choice was made and what …
A research team proposes organizing autonomous AI agents into collaborative "AI research teams" that explore orthogonally to escape local minima, aiming to accelerate scientific discovery. The team applies the approach t…
Sonatype launched Nexus Repository Pro+ to defend against supply-chain attacks that target autonomous AI coding agents installing packages and configuring MCP servers without human review. The launch follows a LiteLLM 1.…
OpenAI retired the gpt-3.5-turbo-0301 model on 2024-09-13, according to the company's deprecations page. The model, a 4k-context version of GPT-3.5 Turbo with a 4k maximum output, is now listed as retired and directs dev…
Microsoft has retired the Azure-hosted Grok 4 Fast (Reasoning) model, listed as grok-4-fast-reasoning, with a retirement date of 2026-05-01, and points users to Grok 4.1 Fast (Reasoning) as the replacement. The retired m…
A developer built an AI citation tracker that scores how well web-grounded models from Gemini and OpenAI identify and describe a tracked entity, and found that a blended visibility score of 29.2 on 21 October concealed w…
Researchers Yewon Kim, Apurva Gandhi, David Chung, Graham Neubig and Chris Donahue released Decomposer, an 8B-parameter model that decompiles symbolic music (MIDI) into executable, editable Strudel programs, detailed in …