
Muse for Mac: Meta Brings Its Personal Agent to the Desktop
Meta shipped Muse for Mac on Sep 18, 2026 — it works across apps, files, calendar, and messages. Here's what's new and the open security questions.

Expert profile
Founder & AI Product Leader
Yash Thakker is a Generative AI expert with over 12 years of experience in product leadership and technical strategy. As the founder of explainx.ai, he has taught over 300,000 learners and built AI platforms serving millions of users globally. He specializes in Agentic AI, Multimodal RAG, and the intersection of LLMs with consumer hardware.

Meta shipped Muse for Mac on Sep 18, 2026 — it works across apps, files, calendar, and messages. Here's what's new and the open security questions.

One week after launch, Meta's Muse reached #1 on the US App Store, with a 4.9-star rating across 13,000+ reviews — alongside the same-week Mac launch.

OpenAI launched Astra for Law on Sep 17, 2026 — GPT-6 Astra plus a 230M-source legal index, beating web-search Astra by 40% on a legal bench.

PrismML shipped Bonsai 2 27B on Sep 17, 2026 — a 5.9GB ternary compression of Qwen3.8 27B retaining 98.2% of full-precision benchmark performance.

Alibaba launched Qwen3.8-Omni-Flash on Sep 18, 2026 — 1M-token omnimodal model, 98% cheaper audio pricing, and agentic video editing via Qwen-MM-Plugins.

Researchers chained a libheif RCE with an OpenAI SSO flaw in under 72 hours — Claude Opus 5 built a working exploit from scratch in 3 hours.

The ten concepts behind every production agent harness — loop, tools, permissions, context budgets, checkpointing, and sub-agent orchestration.

10 concrete things builders are shipping with GPT-6 Astra — agentic browsing, spatial reasoning, code review, and where to learn it hands-on.

Harness engineering is the discipline of building the loop, tools, and recovery logic that turns a raw model into a working agent, explained.

Mosaic cut video tokens 97%, Ponder hit 0.967 F1, Revyl improved accuracy 65% with Gemini agentic video. The architecture behind each result.

AI took 1st, 2nd and 5th in the Metaculus Cup. But FutureEval found human pros still beat the top bots by 1.25 points per question.

Anthropic's CEO proposed $687K salaries for independent AI safety evaluators like METR, aiming to close the pay gap with frontier labs.