A mitmproxy teardown of GitHub Copilot in VS Code revealed that the assistant's context engine shipped a secret from a .env file the user wasn't editing, bundling up to 20 files with 8 edit summaries …
OpenSSH 10.5, released on August 11, just five weeks after 10.4, includes three security fixes and a change in release cadence due to a flood of security reports from AI models or made with AI assista…
Google published a pitch co-signed by Go product lead Cameron Balahan arguing that Go is ideal for AI-assisted software engineering, citing its deterministic toolchain and fast feedback loops. The pos…
Nvidia shipped NeMo Switchyard, an Apache 2.0 Rust proxy that routes LLM traffic between models mid-workflow, alongside the Nemotron 3.5 Lightning model on August 11. The escalation router cut costs 7…
A new paper by Alexander Panfilov, Ilia Shumailov, Luca Beurer-Kellner, Jonas Geiping, Maksym Andriushchenko and colleagues demonstrates that encrypted reasoning traces from proprietary LLM APIs from …
Nvidia Corp. released Nemotron 3.5 Lightning, a 30B-parameter open mixture-of-experts model with 3B active parameters and a 1M-token context window, alongside NeMo Switchyard, an Apache-2.0 model rout…
A new preprint, 'Stealing Reasoning Traces from Proprietary LLM APIs,' reveals that encrypted chain-of-thought blocks from Anthropic, OpenAI, and Google can be replayed into weaker sibling models to e…
Cua's Metal capability shim recovers up to 16.36× llama.cpp throughput in macOS virtual machines by unmasking GPU capabilities that Apple's paravirtualized graphics device underreports. On an M1 Ultra…
Cua, a company building computer-use agent infrastructure on macOS virtual machines, achieved 11–16× faster llama.cpp inference inside Apple Silicon VMs by intercepting Metal capability queries to rep…
NVIDIA released Nemotron 3.5 Lightning, a 30B-parameter mixture-of-experts model with 3B active parameters, and NeMo Switchyard, an open-source routing library, on August 11. The router, which dynamic…
A product manager's guide to organizing Claude Code, created by Adam Faik, went viral on Hacker News, arguing that results with AI agents depend more on file organization than prompting. The guide, wh…
Dan Luu's large-scale evaluations show that the token-efficiency advantage of terse programming languages for AI coding agents largely disappears on real tasks, contradicting a widely cited claim. In …
Salvatore Sanfilippo, creator of Redis, released h3.c, a native C inference engine for MiniMax's open-source H3 video model, rendering video with synchronized stereo audio on Apple Silicon in about 75…
Meta shipped Muse Glimmer, a 30B dense multimodal model from Meta Superintelligence Labs, under the Apache 2.0 license, abandoning the Llama Community License for the first time. The model, designed f…
OpenAI released GPT-5.6-Cyber, a model trained to perform offensive cyber operations with fewer refusals, completing 95% of advanced-cybersecurity requests versus 1.5% for stock GPT-5.6 Sol. The model…
Probing by Shrivu Shankar found that frontier models' actual knowledge can lag their stated cutoffs by months, with Claude Opus 4.7, 4.8, Sonnet 5, and Fable 5 all losing signal around late December 2…
Anthropic now watermarks all text and files generated by its Claude models globally, including API output, with models launched on or after August 2, 2026 embedding an imperceptible statistical waterm…
Ante, a new terminal coding agent from Antigma Labs, ships as a single ~15MB Rust binary with no runtime dependencies and an embedded llama.cpp inference engine, allowing fully offline operation again…
Meta Platforms Inc. released Muse Glimmer, a 30B-parameter dense open-weight AI model, under the permissive Apache 2.0 license, marking its first open-weight release since Muse Spark and a shift from …
A security researcher known as BobDaHacker found that tl;dv, an AI meeting notetaker, left its Firestore 'meetings' collection readable by any authenticated user, exposing 181,874 meeting records from…