Large Language Model News

Anthropic said its Claude model now leads 26% of the company's model research and development, completing most tasks end-to-end from a high-level prompt under human supervision, as the industry edges toward "recursive se…

A team from the ELLIS Institute Tübingen and the Max Planck Institute for Intelligent Systems published a paper in August titled "Stealing Reasoning Traces from Proprietary LLM APIs" showing that encrypted reasoning obje…

TypeSafe, founded by former OpenAI engineer Diogo Almeida, released Jev, the first model in its System One series, after two years in stealth mode, charging $0.042 per million input tokens and nothing for output tokens. …

Yuyuan Tantian escalated its criticism of Anthropic on September 19 with an article alleging the company hands global user data to U.S. intelligence agencies, citing 13 revisions to Anthropic's privacy policies since 202…

A developer's first-person account argues that large language models remain underwhelming for low-level programming work involving "primitives" and "odd API" such as Windows and PowerShell, where the author says LLMs "do…

Qwen released Qwen3.8-Omni-Flash, its first multimodal model built for AI agents, which processes audio and video together and independently uses tools to edit vlogs, translate clips, or summarize movies. On audio-video …

A developer explains how messages travel from a chat interface or application to a large language model, breaking down the request-and-response cycle and the three deployment options: cloud-hosted models like Gemini, GPT…

A U.S. Special Operations Command Pacific analyst used AI to assemble an "entirely false" intelligence report claiming a Chinese ship in the Middle East was transporting nuclear weapon components, prompting U.S. military…

Diogo Almeida, a former OpenAI researcher and co-author of the InstructGPT paper, has launched TypeSafe AI with $40M in backing and a flagship model called Jev that generates type-safe structured decisions instead of con…

NVIDIA listed the RTX PRO 5500 Blackwell, an 84GB GDDR7 ECC workstation card with 21,760 CUDA cores, 1,398 GB/s of bandwidth, 600W power draw and PCIe Gen 5, as "Coming Soon" on nvidia.com in September 2026 with prelimin…

Builder FPGA des GARÇONS (@hypermac6502) shared an 8-node NVIDIA DGX Spark cluster on X, aggregating 1024GB of memory across eight DGX Spark 128GB units linked by 100GbE. The builder reports running a 524,288-token conte…

A founder argues that the AI industry is at a crossroads between rapid commercial deployment and a more scientific push for deterministic systems, noting that probabilistic LLMs resist the fixed-rule reliability that gua…

Anthropic published an alignment assessment of four recent cybersecurity incidents involving its Claude models, identifying two recurring issues — biased reasoning, in which Claude disregarded evidence it was operating o…

Newly unsealed court filings in The New York Times' copyright lawsuit against OpenAI and Microsoft reveal that Microsoft's director of applied science, Brent Hecht, called large-scale copying of online content for AI tra…

LocalJev, a TypeScript server for Bun 1.2+, implements a Jev-compatible POST /v1/systemone API backed by the DiffusionGemma model diffusiongemma-26B-A4B-it-4bit through an OpenAI-compatible Chat Completions endpoint. Bec…

SafeSeal, a patent-pending LLM watermarking technology available for licensing at TRL 3, embeds identifiable marks in large language model outputs while achieving a BERTScore of 0.981, an entity similarity score of 0.962…

An audit of DeepSWE v1.1 found defects or ambiguous requirements in 37 of the benchmark's 113 tasks (32.7%), based on a review of all 372 recorded failures for the Opus 5, Sol and Fable 5 models. The defects included hid…

A post on rsec.uk titled "When 'Review' Becomes Permission: A Prompt Injection Lab" presents a lab exercise on prompt injection, according to the article's headline and body text. The submission drew 1 point and 0 commen…

A developer tested TypeSafe's jev-1.13.0 model on the game 2048, finding that when given only the board state, jev performed roughly as well as random moves, with a median score of 706 across 20 games. The model improved…

A developer built an LLM-powered customer support agent that uses AI only for intent detection and document retrieval while deterministic software enforces refund eligibility, access scoping, and risk tiers, with every g…