Modular announced Mojo 1.0, turning the once‑experimental language into a stable, production‑ready platform. After years of rapid iteration, the release freezes the core syntax, adds Python‑style lambdas, a reliable LSP server, and 1.0‑grade AI Skills, giving the growing community a solid base for long‑term projects. Over 200 contributors helped shape the milestone, signalling a maturing ecosystem.
A new paper shows that Anthropic, OpenAI and Google return encrypted reasoning blocks that can be swapped into weaker models, forcing them to output the original chain‑of‑thought in plaintext. The authors decoded 315k public traces, exposing 367 PII records, 182 credentials and enabling hidden prompt‑injection attacks, effectively stealing proprietary model reasoning.
Anthropic now accounts for 43.5% of U.S. firms’ paid AI spend, nudging OpenAI to 39.7% and outpacing its own Fable 5 rollout. The latest Ramp AI Index shows slower adoption of Anthropic’s new model and growing interest in cheaper open‑source alternatives, signalling a shift in corporate AI buying patterns.
In a single weekend the author crawled 560k maker‑oriented domains, stored only a kilobyte of metadata per site, and powered a fuzzy search UI using an overnight GPU for roughly $10. The result is a personal index that surfaces indie projects, art and hardware without the corporate noise of traditional search.
Simon Willison’s new alchemy-utils library ships its first alpha, re‑implementing sqlite-utils on top of SQLAlchemy. It lets you run the same CLI commands against SQLite, DuckDB, PostgreSQL and other backends, turning a SQLite‑only tool into a multi‑database workhorse.
Google’s new JAX‑native library Tunix removes the classic TPU‑idle bottleneck in multi‑turn, tool‑using LLM agents by streaming asynchronous rollouts through a producer‑consumer pipeline. The framework keeps accelerators busy, adds low‑overhead RL metrics, and lifts training throughput by up to several×.
MCP-Memory is an open-source server that adds fast, persistent memory to AI agents like Claude Desktop or Cursor. It stores records in the Open Knowledge Format and indexes them with SQLite FTS5, delivering sub‑20 ms lookups and full‑text search. The zero‑boilerplate setup lets developers add long‑term context without rewriting agents.
A deep dive reveals that any scheme embedding hidden signatures in LLM output can be stripped with trivial edits, making EU-mandated watermarks ineffective. This means regulators and developers must look beyond watermarking to reliably flag AI‑generated text, or risk non‑compliance and false positives.
Delta adds a multiplayer IDE where code, git history, and chat stay in sync via DeltaDB. Teams and AI agents can comment on live code, preserving context as it evolves, cutting the guesswork of diffs and speeding up reviews. It works with existing repos and syncs locally and in the cloud.
Nine PBS filed a lawsuit against Iron Mountain after its cloud‑storage vendor went defunct, blocking access to over 50 TB of archival material spanning 70 years. The station argues it owns the data and seeks a court order to prevent deletion and retrieve the files, underscoring the preservation risks of third‑party cloud services.
The skitter‑creek‑bath‑salts repo shows a low‑level hack that rewires DRAM address translation on AMD Family 16h CPUs, letting attackers reach protected memory, unlock the platform security processor, SMM, and even microcode updates. By “spaghettifying” DRAM, it demonstrates that hardware‑level isolation can be broken, raising serious firmware security concerns.
Echo partnered with NanoClaw to harden its container images, running multiple scanners, upgrading safe libraries, backporting patches, and contributing fixes upstream, cutting the CVE count from roughly 1,400 to under 800. The process shows a reproducible path for securing AI‑focused containers and lowers supply‑chain risk.
Over three years, Yelp transformed a monorepo of more than a million lines from Flow to TypeScript by converting one package at a time, keeping both type systems running side‑by‑side in production. Using the open‑source flowts tool plus custom codemods, they avoided massive freezes, preserved type‑checking trust, and saved tens of thousands of engineer hours.
You’re caught up.
That was all 13 stories in the 13 August issue.