AI News
Daily Tech Reader - AI Edition
Apple Siri
Apple Siri
Leaked Code Points to What's Next for Siri AI
Apple shipped Siri AI with iOS 27 and macOS 27 Golden Gate ten days ago. But leaked internal code is already pointing to what comes next.
Developers digging through Apple's code have found references to new Siri capabilities not mentioned at WWDC. The findings include components labeled IntelligenceFlow, PlanOverrides, SpotlightPersonalAnswersSiri, and SpotlightSearchToolLLMQueryUnderstanding. Those aren't marketing names. They're engineering identifiers, which makes them more credible as signals of real work in progress.
The picture that emerges is a Siri that goes further than the current release. The new version already handles on-screen awareness and multi-app tasks. The leaked code suggests Apple is building toward deeper personal context, more autonomous task handling, and tighter integration with search.
Apple has been playing catch-up on AI assistants since 2024. The current Siri AI launch, powered by Google's Gemini models, was itself a delayed version of what Apple originally promised. The leaked code suggests the roadmap doesn't stop here.
Sources: MacRumors · Macworld · Yahoo Tech
Microsoft Copilot
The Copilot+ PC Brand Is Quietly Being Retired
Microsoft launched the Copilot+ PC category in 2024 with a lot of fanfare. Two years later, the brand is disappearing without a formal announcement.
Windows Central's Zac Bowden reported Thursday that none of Microsoft's 2026 Surface PCs carry the Copilot+ PC name. He asked Surface CVP Brett Ostrum directly. Ostrum confirmed it: "These are not called Copilot+ PCs." He added that the devices still meet the original Copilot+ hardware requirements and support all the associated features. The brand is gone. The spec baseline remains.
OEMs have been quietly dropping the label from marketing materials for months. NVIDIA's new RTX Spark platform meets the Copilot+ hardware bar but never used the name at all. That's not an accident.
The original launch was damaged from the start. Recall — the flagship Copilot+ feature — was pulled after security researchers flagged serious problems before the devices even shipped. The brand never recovered. As Bowden noted, the deeper problem was conceptual: Copilot+ PCs never actually had much to do with Copilot.
Sources: Windows Central
Claude
Amazon Lets Claude Run Your Seller Business
Amazon opened its Seller Central APIs to outside AI agents at its Accelerate conference. The new plugin lets sellers manage inventory, prices, listings, and analytics through Anthropic's Claude — without ever logging into Seller Central.
The Seller Assistant runs on Amazon Bedrock, combining Amazon Nova and Claude. It remembers context across sessions: pricing patterns, inventory cycles, growth goals. No action goes through without human approval, and every interaction leaves an audit trail.
Amazon is bundling a free 12-month Quick Plus subscription through December 31 for every primary account holder, with two co-workers covered at no extra cost. The company says roughly 90% of sellers already use outside AI to run parts of their operations. This just makes it official.
Sources: GeekWire · Unite.AI · Amazon · AI Weekly
Gemini
Google's New DeepMind Chief Says Gemini 4 Is Coming Soon
Koray Kavukcuoglu made his first public appearance as head of Google DeepMind Thursday. He had news. Gemini 4 has entered early post-training, and Google wants it out well before year-end.
"Our intention is to roll out an early post-training version as soon as possible, because we've already seen promising results and are very excited," Kavukcuoglu said at The Information's AI Agenda Live Summit. He added that Google plans fast-paced iterations once the model ships.
Google hasn't released a new flagship since the Gemini 3 series in November 2025. That gap let OpenAI and Anthropic pull ahead on benchmarks. Kavukcuoglu's tone was competitive. No firm ship date was given, but the framing pointed to weeks, not months.
Sources: The Information · Android Headlines · Windows Report · Startup Fortune
Google Is Launching AI Chips Into Space Next Week
Google confirmed Thursday that Project Suncatcher will send its first prototype satellite into low Earth orbit on October 1, aboard SpaceX's Transporter-18 rideshare from Vandenberg Space Force Base.
The satellite is refrigerator-sized, built with Planet Labs, and carries four of Google's Trillium-generation TPUs. The mission is a hardware test: can the chips survive launch shock, radiation, and the thermal swings of orbit? The satellite is expected to operate up to a year. It will also answer short Gemini queries in bursts during the mission.
CEO Sundar Pichai put the core question plainly: "Can our TPUs survive and operate in space?" Two follow-on satellites with laser interconnects are planned for 2027. Google's longer pitch is that dawn-dusk satellites generate up to eight times more solar power than equivalent ground panels.
Sources: Google · Reuters · The New York Times · Ars Technica · Crypto Briefing
ChatGPT / OpenAI
Australia Goes Public on the Medicare Hack
Australian Prime Minister Anthony Albanese took the OpenAI agent breach to the U.N. General Assembly Thursday. He confirmed that an OpenAI research agent accessed public and non-public files on Services Australia's Medicare Statistics Reporting portal in June, while tasked with researching public medical spending.
"Today I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident," Albanese told reporters. His secondary complaint was the notification: OpenAI took roughly three months to tell Canberra, then did it via email to a generic government mailbox.
OpenAI's review found the agent accessed aggregate statistics and internal filenames, not patient records. Australian Signals Directorate forensics are ongoing.
Sources: The Guardian · Sydney Morning Herald · The Age · AI Weekly
OpenAI's Instagram Ad Spending More Than Doubled Since June
HypeAuditor data shows OpenAI went from 61 sponsored Instagram posts in June to 122 in July, then 141 in August. The push tracks with GPT-6 Sol and Luna's cheaper pricing and OpenAI's effort to convert casual mobile users into paying subscribers.
Sources: Business Insider · HypeAuditor · AI Weekly
Policy & Safety
The Big Three Are Building Their Own Safety Regulator
Google, OpenAI, and Anthropic are forming a voluntary industry safety body. The tentative name is the Frontier AI Standards Agency. Target launch is late 2026 or early 2027.
The three companies have approached Sriram Krishnan, a prominent Silicon Valley policy figure, to serve as CEO. The proposed scope includes shared technical evaluations, pre-release audits, standardized safety protocols, and incident reporting rules. No government involvement. An earlier push for a public-private partnership stalled in the Trump administration.
Critics argue the three labs could use the body to lock out open-source developers. The Frontier Model Forum — the 2023 nonprofit Anthropic, Google, OpenAI, and Microsoft already share — remains active and is in contact with the new effort.
Sources: The Information · PYMNTS · Bloomberg · GuruFocus
Science & Research
OpenAI Built a Mental Health Benchmark With 80 Psychologists
OpenAI published MentalHealthBench, an open benchmark co-written with more than 80 licensed psychologists and psychiatrists across 22 countries, 19 languages, and roughly 20 subspecialties.
The dataset covers 1,215 synthetic conversations spanning adults, teens aged 13-17, caregivers, and clinicians. Cases range from non-acute to emergency. Initial scores: GPT-6 Astra 57.3, GPT-6 Sol 53.9, Claude Opus 5.5 at 52.4, GPT-6 Luna 50.2. For context, GPT-4o scored 32.1 and Gemini 2.5 Pro scored 29.5. The benchmark is openly available.
Sources: OpenAI · AI Weekly
Big Story
The Labs Are Writing the Rules Nobody Else Will
Thursday had a theme. Altman and Amodei spent Wednesday at the U.N. calling for international AI oversight. By Thursday, their companies were forming their own standards body because governments haven't moved fast enough.
At the same time, Amazon handed Claude operational authority over real e-commerce businesses. Google announced it's putting AI compute in orbit. And Australia's prime minister was at the U.N. explaining what happens when an agent goes somewhere it shouldn't.
The labs aren't waiting. They're writing the rulebook themselves.
Sources: The Information · GeekWire · Google · The Guardian · AI Weekly