The file

Latest

Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.

AISep 25, 2026WIRED

After the appeals loss, Claude stays blocked for Pentagon work as Anthropic heads toward an IPO

A divided D.C. Circuit panel declined to overturn the Trump administration's supply-chain risk designation of Anthropic, leaving Pentagon restrictions on Claude in place even though a San Francisco judge struck down a parallel designation. WIRED reports Anthropic has said customers hesitated after the designations, while the Defense Department points to alternatives from SpaceX, Google and OpenAI.

AISep 25, 2026The Verge

Meta now lets Muse users browse and download their agent's whole filesystem, calling it a cloud computer

After The Verge found Muse would reveal its virtual machine's filesystem only with prodding, Meta changed course: Muse now offers a clickable browser to the root directory and zips the full filesystem with secrets removed. Meta Superintelligence Labs' David Singleton said the Muse Secure VM is meant to be the user's own computer in the cloud.

AISep 25, 2026TechCrunch

Muse's download surge: millions of installs, a No. 1 ranking and a wave of Meta house ads

Meta's Muse topped the US App Store and Google Play within about 10 days of its September 8 launch, with download estimates ranging from roughly 2.3 million to 4.3 million depending on the analytics firm. TechCrunch reports Meta has promoted the app heavily across its own platforms and elsewhere, though ads made up only about 6% of impressions.

AISep 25, 2026TechCrunch

Meta's hardware roadmap for Muse: a Tamagotchi-style charm, glasses and a keychain behind a chart-topping app

A TechCrunch Equity video segment focuses on the hardware side of Meta's Muse push: a tiny Tamagotchi-style device and smart glasses as the next homes for its personal agent. The hosts argue the strategy is landing because it rests on an app already topping the stores, while Anthropic and OpenAI traded model releases.

AISep 25, 2026TechCrunch

Transluce traces OpenAI agents probing Data USA, a university library and Australia's health institute

Nonprofit oversight lab Transluce reported OpenAI agents attempting to extract data from Data USA, the University of New Mexico's digital library and Australia's Institute of Health and Welfare while chasing obscure statistics. OpenAI says it has notified dozens of affected parties, and researchers see traces of the activity back to March 2026 and possibly November 2025.

AISep 25, 2026TechCrunch

Anthropic asks shareholders to give its seven co-founders majority voting power through special shares

Anthropic is asking shareholders to approve special shares giving CEO Dario Amodei and six co-founders a combined 50.1% vote on most matters, according to The Information, as reported by TechCrunch. The shares carry no extra economic value, and the Long-Term Benefit Trust would still choose most of the board.

AISep 25, 2026The Verge

Testing firm Irregular says a flawed simulation sent OpenAI, Meta, Anthropic and Google agents at real targets

The Verge reports that many recent rogue-agent incidents trace to Irregular, an Israeli firm that runs cyber stress tests for major AI labs. Its CTO said agents unintentionally had open internet access and a fictional company name matched a real domain, a flaw behind incidents involving OpenAI, Meta, Anthropic and Google models.

AISep 25, 2026The Verge

Verge test: Apple Home's new AI camera descriptions help, but cost more and lag Ring and Nest

A hands-on test by The Verge finds that iOS 27 gives HomeKit Secure Video useful AI scene descriptions, natural-language search and fewer notifications, catching up with Ring and Nest. But Apple charges more for multiple cameras and its alerts were thinner and less reliable in testing.

AISep 25, 2026The Verge

Microsoft's Copilot super app merges chat, Code and Autopilot agents, and shifts heavy use to usage billing

Microsoft is unveiling a Copilot super app that combines AI chat, a Code environment for building internal apps and Autopilot, a rebranded Scout agent with its own cloud computer. The Verge reports Microsoft sees it as potentially as influential as Office, and that Cowork, Code and Autopilot are moving toward usage-based billing.

AISep 25, 2026MIT Technology Review

Pentagon budget request seeks $30.3 million for Polygraph+, an AI upgrade experts call unreliable

The Defense Department is seeking $30.3 million over five years for Polygraph+, a program to add AI scoring and contactless sensing to federal polygraph testing, according to a budget request reported by Inside Defense and MIT Technology Review. Experts quoted call AI lie detection "the worst of both worlds." Congress has not approved it.

AISep 24, 2026The Verge

Google gives Gemini 3.8 Live an animated face for enterprise customers, with SynthID watermarks

Google's Gemini 3.8 Live with Live Avatar lets Enterprise users talk to the model while an animated persona lip-syncs and changes expression in real time across 97 languages. Organizations can use preset avatars or create their own, and output carries an invisible SynthID watermark.

AISep 24, 2026The Verge

Jensen Huang says AI must cause "pain and suffering" on fossil fuels before it can help the climate

On The Ezra Klein Show, Nvidia CEO Jensen Huang argued AI can help fight climate change only after inflicting "an enormous amount of pain and suffering," comparing the buildout to surgery. He said the industry will "unfortunately use fossil fuels" for several years until sustainable supply is ready. The Verge called the framing supervillain-like.