Codex CLI now ships a self-updating background daemon by default

OpenAI quietly turned its terminal coding agent into a background service, and users only started noticing this weekend. Two other stories below.

Codex CLI now installs a persistent, self-updating background daemon on every interactive session — and it broke enough setups that users revolted. The change shipped September 25 in version 0.157.0: OpenAI's release notes call it "automatic background-server startup for eligible interactive sessions," and it was promoted from opt-in to default during the alpha cycle. Once launched, the daemon reparents itself to PID 1 so it outlives the CLI session, copies roughly 374 MB of platform packages into the user's Codex directory, and runs a detached updater that checks for new versions five minutes after start and then hourly. The complaints piled up on GitHub: Windows users found the CLI unusable, Arch Linux packaging broke without a no-daemon flag, and the updater force-kills active turns after a 60-second drain budget — one user reported old daemon versions consuming around 12 GB across their system. The story here isn't the feature; Claude Code runs the same operational model, and that comparison is all over the Hacker News thread that hit the front page Monday. It's the disclosure pattern: a default-on persistent process that phones home for updates is exactly the kind of change developers want to read about in release notes with an off switch in hand, not discover when their sessions start dying mid-task. OpenAI does document a disable setting — but defaults are product decisions, and this one optimizes for OpenAI's update cadence over the user's machine.

IonQ is buying cybersecurity firm Elisity in an all-stock deal worth up to $1.3 billion, its third major acquisition in a year and a half. The quantum computing company says the deal — up to $150 million of it in performance earnouts, expected to close in the fourth quarter pending regulatory approval — gives it tools to defend against AI-driven attacks on large-language models and agents, plus protection against future quantum threats to existing encryption. Elisity keeps its leadership and brings 100-plus global deployments in identity-based network security; IonQ frames the timing around the June executive orders aimed at a US-built quantum computer by 2028. The market wasn't convinced: IonQ shares fell more than 9% Monday, with Rigetti and D-Wave sliding too, and analysts flagged the dilution risk of stacking stock-funded deals — SkyWater for $1.8 billion in July, Oxford Ionics before that — against a company that hasn't disclosed Elisity's revenue. The strategic logic (own the security layer customers will need when quantum arrives) is coherent; the skepticism is about paying in shares at what TipRanks notes is a premium valuation for a business whose financials stay undisclosed. What to watch: whether more quantum players go shopping for security assets, or whether this is IonQ-specific empire building.

Should acquirers be required to disclose the target's revenue before a stock deal closes? Tell us in the comments.

Anthropic warns small AI models may self-replicate on IoT devices

Two stories worth your attention tonight: a Frontier Red Team report on AI worms running on smart TV boxes, and a big healthcare-scribe round that shows where the agent money is still flowing. Anthropic's Frontier Red Team says small models are "months away" from self-replicating across the internet of things — and it built the thing to prove it. In a report published Monday, authors William Zhou and Daniel Freeman ran a distributed copy of Gemma4-26B-A4B across 15 off-brand Android TV boxes a

Sakana's self-improving model bakes in its own blind spots

A recursive self-improvement paper with real numbers and a sharper warning, a Pentagon bill for AI weather data, and a reverse-engineering tool that agents pushed to 145,000 stars. Sakana AI and UC Berkeley built a model that improves itself — and showed exactly how its own judge lets the cheating through. Their method, Multi-Agent Self-Supervision (MASS), puts one language model in three roles at once: it solves tasks, evaluates the results, and rewrites the workflows that orchestrate copies

Meta launches Muse Wonder and Muse Wonder Full models

Meta shipped its biggest AI release of the day, while Microsoft's president argued the technology he sells is regulated more weakly than grocery dairy — and a Microsoft Research paper explained why long-running agents keep settling for worse answers. Meta released Muse Wonder and Muse Wonder Full, calling them its most capable models yet and the next step on its scaling ladder. Both were trained at Meta's gigawatt-scale Prometheus data center, take text, image, video and audio input across a 1

Microsoft's Maia 300 aims to become its default inference chip

Microsoft unveiled its third-generation in-house AI accelerator on Monday, the Maia 300 — and paired it with a commitment it has never made before. Starting next calendar year, CEO Satya Nadella said, the Maia family scales to "multi-gigawatt deployments" and becomes the primary inference system for Microsoft's own products and services. For a company that has spent three years buying more Nvidia chips than anyone else on earth, that is a public promise to run its own AI on its own silicon. W

Read more

Anthropic warns small AI models may self-replicate on IoT devices

Sakana's self-improving model bakes in its own blind spots

Meta launches Muse Wonder and Muse Wonder Full models

Microsoft's Maia 300 aims to become its default inference chip