Moonshot AI released the full weights for Kimi K3, a 2.8-trillion-parameter open-weight model that is the largest ever released, nearly triple the size of its predecessor Kimi K2. The model validates …
Anthropic's Claude Mythos Preview model found improved attacks against the HAWK post-quantum signature scheme and a reduced-round AES, but verifying the results took researchers nearly a month, highli…
Anthropic's Claude found weaknesses in reduced-round ciphers and the NIST post-quantum candidate HAWK, but full-strength AES-128 remains secure. Each attack cost about $100,000 in API spend and target…
A tutorial by Rachel Goldstein shows how to build a browser-based voice assistant using the OpenAI Realtime API and WebRTC, with a Node.js server that mints ephemeral tokens and a static page that str…
Moonshot AI's Kimi Linear attention architecture, introduced in October, now powers the company's 2.8-trillion-parameter Kimi K3 flagship model released in mid-July, marking the first production deplo…
A Lean 4 project by Peter Schilde, verified-3d-mesh-intersection, uses AI agents to write over 1,000 lines of implementation and more than 60,000 lines of formal proofs, leaving humans to review only …
Google security engineers Joseph Valente and Michal Zalewski argue in a new ACM Queue paper that zero-trust security models are obsolete for the AI era, proposing a replacement called Beyond Zero that…
Apple shipped macOS Tahoe 26.6 on July 27 with fixes for 155 CVEs, one of its largest single security drops, with no zero-days flagged as actively exploited. The bulletin reveals AI systems are now ro…
Fermisense, an AI consultancy, reports that a 9B Qwen specialist fine-tuned for $500 on two RTX PRO 6000 GPUs over three and a half days scored 87.3% on a simulated catalog-review benchmark, outperfor…
Fermisense, a consultancy, spent roughly $500 of GPU time fine-tuning a Qwen3.5-9B open model with reinforcement learning and claims it beat GPT-5.5, Gemini 3.1 Pro, Claude Opus 4.8, and Claude Fable …
Claude Opus 5 passed 24% of 17 checkpoints on a subset of SlopCodeBench, quadrupling the 6% pass rates of Opus 4.8 and Sonnet 5, but achieved this by producing roughly 29,000 source lines versus about…
Anthropic CEO Dario Amodei published a position on open-weights AI models on July 27 that opposes a recent open letter signed by Meta, Microsoft, IBM, and others, arguing that open-weights models pose…
Feyn Labs released FeyNoBg, a 263M-parameter background-removal model that leads on four of eight standard benchmarks and stays within 2% of the leader on the rest, but the weights are licensed CC-BY-…
Microsoft's in-house MAI-Code-1-Flash model, post-trained within GitHub Copilot's harness, achieves approximately 10% higher code accept rate than GPT 5.4 Mini and Claude Haiku 4.5 in VS Code, with 6%…
GitHub's July 14 changelog announced AI security detections for Shell, Dockerfiles, Terraform, and PHP on pull requests, but access requires a GitHub Code Security (Advanced Security) subscription, a …
Bun's rewrite from Zig to Rust took 11 days using Claude agents, but shipping the resulting code has taken months, with no v1.4.0 release after 11 weeks and a backlog of 2,475 open pull requests from …
Netflix's AI Platform team published a detailed account of its LLM serving platform, revealing that version pinning between NVIDIA Triton Inference Server and vLLM, a Python GIL bottleneck, and KV-cac…
Moonshot AI released the weights for Kimi K3, a 2.8-trillion-parameter mixture-of-experts model, making it the largest open-weight model ever. The model activates 50 billion parameters per token and a…
Vercel Labs released Scriptc, a TypeScript-to-native compiler that produces executables without a JavaScript engine, with hello-world binaries around 200–320KB and single-digit millisecond startup tim…
A 780 KB Rust proxy called cursor-bridge lets users run Anthropic's Claude Code against Cursor's backend, billed to their Cursor subscription instead of a separate Claude plan. The tool, which hit Hac…