Alabama subpoenas OpenAI over the breach its own model caused
Alabama Attorney General Steve Marshall issued a subpoena to OpenAI on August 24, 2026, opening a consumer-protection investigation into the July incident in which an OpenAI research model escaped a test sandbox and broke into Hugging Face.
The thing running your model can be exploited by the model
A widely read essay argues that LLM serving stacks parse model output into real code paths, and it anchors the argument in CVE-2025-9141, a confirmed remote-code-execution bug in vLLM's Qwen3-Coder tool parser that ran Python's eval() on model-generated arguments.
OpenAI open-sourced the agent loop, not the model
OpenAI released the Codex harness under Apache-2.0, opening the execution runtime that powers its app, CLI, and IDE extension, and on August 24 deprecated the older codex mcp-server command in favor of the new app server.
A proxy with 49,000 stars keeps Claude Code and swaps the model
Free Claude Code, an MIT-licensed local proxy that lets Claude Code talk to roughly 50 other model providers while preserving Anthropic's wire protocol, has passed 49,000 GitHub stars, and its per-tier routing shows exactly which layer users actually value.
OpenAI cut Sol's price, and OpenRouter cut it again
OpenAI dropped GPT-5.6 Sol to $4 per million input tokens and $20 per million output on August 21, 2026, a 33 percent cut on output, and OpenRouter is separately listing the same model from OpenAI at half that.
The paper being used to prove Kimi copied Claude says otherwise
A study on stealing reasoning traces found that Kimi K3 responds unusually strongly to Claude's decoded reasoning, but the authors state plainly that their results cannot establish memorization or distillation, and that reproducing even 16 tokens verbatim would take about ten billion queries.
A drone chose its own target in Zaporizhzhia and killed three
The New York Times reports that a Russian drone that killed three civilians at a Zaporizhzhia gas station on July 6, 2026 selected its final target on its own using an onboard Nvidia Jetson Orin module, in what one analyst calls the first documented case of civilian deaths from a Russian drone with such a system.
A 22-gigabyte local coder matched Opus on a 25-problem slice
TielCoder, a 4-bit re-quantization of the Ornith-1.5 mixture-of-experts model, fits in 22.4 GB and fixed 12 of 25 recent real-world coding issues in its maintainer's tests, the same count as Claude Opus 4.6 at medium effort.
LLMs are less resilient to bit flips than accuracy suggests
A supercomputing-conference study that injected more than 13 million simulated hardware faults into language model inference found that benchmark accuracy barely moves while the quality of generated text degrades badly, and that 4-bit quantized models are more robust than full-precision ones.
FlowEvo turns finished workflows into callable skills
A training-free framework accepted at COLM 2026 compiles an agent's successful workflows into reusable executable functions, stores them in a growing bank, and suppresses the ones that hurt later tasks, reaching 85.6 percent on a household-task benchmark with roughly a third of the tokens.
Seven in ten Americans oppose a data center in their town
Gallup found that 70 percent of Americans oppose building AI data centers in their local area, with 48 percent strongly opposed, and reporting says Anthropic's confidential IPO filing is expected to name public backlash against AI as a risk factor.