Granola Ships Dictation for Mac, Leaves iOS and Windows as Promises
Granola shipped a free, unlimited Mac dictation feature that turns speech into text in any app, answering Wispr Flow's meeting-notes pivot and Fireflies' dictation launch.
We measure the distance between what was announced and what has been confirmed, using primary records and their conditions.
Latest
Pennsylvania's new AI ad law reaches from Super Bowl spots to wing-special flyers
Pennsylvania Governor Josh Shapiro signed a law requiring disclosure labels on AI-generated or significantly AI-modified ads, covering everything from national TV spots to local restaurant flyers.
Google's October 16th Privacy Policy Change Opens Gmail and Photos Data to Ad Targeting
Google quietly added a clause to its privacy policy on October 16th letting Gmail, Drive, and Photos data feed personalized ads if those services are linked to AI features like AI Mode or Gemini.
Apple's New App Store Badge for the iPhone Duo Is a Software Label, Not a Hardware Certification
Apple's "Made for iPhone Duo" badge just requires building in Xcode 27.1, no fee or lab testing, but developers say even well-designed apps need real rework for the foldable's two screens.
Nirva's $249 Pendant Promises to Listen Only to You. It Hasn't Shipped Yet.
Nirva has raised $33M across a seed and new pre-Series A for a $249 titanium pendant that claims to record only the wearer's voice, but the no-retention privacy promise is still an unverified company claim, and pre-orders start before any regulatory clarity on wearable wiretap rules.
Claude Duet Arrives as a Research Preview, Not a Finished Agent
Anthropic's new Claude Duet agent, built on Claude Opus 5.5, ships as a research preview on iOS, Android and the web, not a finished product.
Nvidia's OpenShell Pitch Leans on Hardware Lock-in and a Telemetry Opt-Out, Not New Rules
Nvidia's new OpenShell Reference Architecture promises to govern unruly AI agents, but the partners it names are mostly evaluating or building toward it, not bound to it, and the runtime still ships with open security issues and default telemetry collection.
FairStress Finds Language Models Abandon Correct Answers Unevenly, Not in the Direction You'd Expect
Ai2's FairStress tests whether language models keep a correct decision under rhetorical pressure, finding models flip away from correct answers 6.9 to 9.3 points more often when that flip favors a minority-coded candidate.
FCA to extend incident reporting rules to AI near-misses, ahead of Parliament's own plan
The FCA says firms must now report significant AI failures and near-misses under its existing incident rules, even as Parliament weighs separate statutory AI reporting duties.
Four European Carriers Eye Joint Spectrum Bid for Satellite Service, but the Hardware Partner Is Still Missing
Deutsche Telekom, Orange, Telefónica, and Vodafone signed a non-binding letter of intent to jointly bid for spectrum for a European direct-to-device satellite service rivaling Starlink, but have not named a satellite partner.
How Character.AI's Bankruptcy Let Disney Buy the Company Without Buying Its Lawsuits
Character.AI's bankruptcy let Disney absorb its staff, tech, and former CEO while leaving abuse lawsuits and state investigations with a stripped-down shell company.
Nvidia's RTX Spark Laptops Are Fast at Local AI, If You Buy the Right Configuration
Nvidia's RTX Spark chip, debuting in HP's OmniBook Ultra 16 and rivals, beats Apple's M5 Max and AMD's Halo on local AI inference, but memory bandwidth, RAM tier and price (up to $7,000) decide if it's worth it.
Meta Strikes AI-Training Clause From Commercial Terms After Legal Pushback
Meta deleted language from its commercial terms that let it mine content from any domain, web page or URL to train its AI, after a law professor flagged the clause as covering news publishers.
