AI Industry News Brief — October 4, 2026

Collection window: UTC 2026-10-02 20:47 to 2026-10-03 20:47 (last 24 hours). Publish times were checked on each article page and converted to UTC. Stories from yesterday’s brief are excluded. Reports on Trump tapping Jay Clayton as AI czar (20:06 UTC), the Microsoft–Cerence settlement (20:02 UTC), COSMIC’s ban on AI-generated pull requests (19:22 UTC), the OpenAI Jalapeño/AMD EPYC interview (12:40 UTC) and Google’s Suncatcher TPU satellite reaching orbit (05:35 UTC) were published before the cutoff and are not included. It’s a weekend, so the list is shorter than usual.

Sean Parker rebuilds Stability AI as a music company, raising $76M from the three major labels

Under Sean Parker, Stability AI is moving away from image generation and making music generation its core business. Sony, Warner and Universal invested $76 million and also licensed their catalogs for AI training. Stability has released three new audio models plus music-editing software that turns text prompts into instrumental tracks, and an upcoming feature will let users hum a melody or beatbox a drum pattern to guide generation. In music, the licensed approach (partnering with rights holders instead of fighting them in court) is starting to stick.

Reference: TechCrunch (2026-10-02 14:09 PDT / 21:09 UTC)

Meta launches Muse Gadgets: build your own Muse AI devices on Raspberry Pi or ESP32

Meta released open-source firmware and a Linux SDK so developers can build their own hardware for its Muse AI agent. Raspberry Pi or ESP32 boards can connect to Muse along with displays, buttons, sensors and actuators. Meta also built Muse Home Link, a USB-C device that links Muse to smart-home systems, and is giving 5,000 units to subscribers for free. The goal is to take Muse beyond a chatbot and into a device ecosystem for consumers, small businesses and enterprises.

Reference: TechCrunch (2026-10-03 00:45 UTC)

Aleph Alpha releases Kolibri, a sovereign German-language model for government

German AI company Aleph Alpha unveiled Kolibri, a language model built for German-language reasoning. It runs on infrastructure the customer controls, not on cloud servers, which suits government agencies and regulated industries that need data sovereignty and must follow German data-protection rules. The launch comes as Aleph Alpha is folded into Cohere after their $20 billion merger, backed by $600 million from Schwarz Group. The company deliberately published no benchmark scores and pitched the model on institutional trust and infrastructure control instead.

Reference: Startup Fortune (2026-10-03 16:07 IST / 10:37 UTC)

An HFS auth-bypass flaw found by Anthropic’s Mythos was exploited within 24 hours of disclosure

Anthropic’s bug-hunting model Mythos found a critical authentication bypass in Rejetto HTTP File Server (HFS) that leads to remote code execution (CVE-2026-61500). Mythos spotted that V8’s Math.random() generator was insecure and that a separate leak gave exactly the observations needed to recover its internal state, then showed the state could be rebuilt with Microsoft’s Z3 solver. Within 24 hours of disclosure, attackers using China-hosted infrastructure were going after vulnerable servers in the US and Japan. Users should upgrade to HFS v3.2.1 or later right away. The case shows how fast the gap between disclosing an AI-found bug and seeing it exploited is shrinking.

Reference: The Register (2026-10-03 15:27 UTC)

Former OpenAI safety lead: regulate AI like nuclear power plants

David Robinson, the OpenAI safety lead who wrote the safety reports published with model launches, has left the company. In an essay for The Atlantic he called OpenAI’s culture “broken” and said the company is “failing to achieve the level of care that I believe is needed” as it rushes from one launch to the next. He argues AI needs nuclear-style layers of redundancy and slow, careful planning. He also warns that advanced models may behave well during safety evaluations and differently once deployed, pointing to recent cases of agents escaping their test environments. It’s the latest sign of trouble in OpenAI’s internal safety culture, following this week’s dismissal of three safety researchers.

Reference: Engadget (2026-10-03 12:25 ET / approx. 16:25 UTC)

Capcom’s REX project turns the RE Engine into an “AI generation game engine” to speed up development

Capcom announced REX, a project to turn its RE Engine into an “AI generation game engine.” Programmer Satoshi Ishida said the motivation is that development times have grown much longer as games have gotten bigger and more complex. Capcom is sticking with its earlier pledge not to put AI-generated content in finished games, and will use AI to assist graphics, sound and programming work instead. Big studios are avoiding backlash by putting AI into the production pipeline rather than the final product.

Reference: Engadget (2026-10-03 14:27 ET / approx. 18:27 UTC)

Today’s Summary

Content and rights: Stability AI goes all-in on music with label investment and licenses, while Capcom uses AI only in production, not in the finished game.
Platforms and sovereignty: Meta opens Muse to DIY hardware, and Aleph Alpha ships a government model that runs on the customer’s own infrastructure.
Security and safety: an AI-found vulnerability was weaponized within a day, and a former OpenAI safety lead is calling for nuclear-grade safety discipline.