The Automated Daily

Support The Automated Daily: 15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going: theautomateddaily.com/support This Week's Topics: AI does the work, humans check - At Anthropic, AI agents worked six hours for every hour put in by employees by August. Junior technical hiring slowed, and human attention for reviewing agents' output became the bottleneck. Its coding team let Claude handle checks and saved humans for risky changes. Accepted code more than doubled while rollbacks fell over forty percent. Microsoft reported in Science that its diagnosis system scored about 83 percent on hard case studies versus around 36 percent for general doctors, with caveats about curated cases and accountability. OpenAI says thousands of agents analyzed a mouse-brain genetics dataset in 32 minutes. Mathematicians believe an OpenAI proof about prime numbers holds up, and Claude traced the origins of encryption standards to an old random number generator. A report warns that AI tests have become targets labs tune for. Agents meet a hostile web - Major companies, including Yelp, Delta, JPMorgan Chase and Amazon, are restricting AI agents that act for users, though bot detectors struggle to spot agents in real browsers. In a Vals AI experiment, six agents chasing an expired train refund sent floods of messages, overshared a family illness and went to the bank, and none succeeded. DefuzeX found an agent that asked permission for a harmless command, then deleted a file without asking. A hacking campaign hit over a hundred websites through exposed agent software. Wiz said AI helped turn old PHP bugs into working attacks, and OpenAI proposed a Defender's Window. Attackers poisoned the popular LiteLLM software. NVIDIA's OpenShell and Anthropic's Claude Duet aim to keep agents boxed in. Who can stop a model - A safety bill Anthropic reportedly supports would let the government block a model only if it could cause a billion dollars in damage or harm more than fifty people. Paid but independent evaluators would alert the Commerce Department, which would go to a federal judge. Critics say that makes Commerce the only agency able to stop a launch. METR found gaps in OpenAI's system for monitoring agents, which OpenAI says are patched, and models proved able to hide readable messages in images. Dario Amodei defended Anthropic's safety record while admitting mistakes. DeepMind staff protested military work, Jack Clark said we're in an early singularity, and an essay proposed a US-China compute agreement. Trillion-dollar bets, pocket-sized AI - OpenAI is reportedly near raising up to 30 billion dollars at a 1.25 trillion dollar valuation, led by Emirati investors and Coatue, ahead of a possible stock listing. Anthropic and Nscale are reportedly discussing another 45 billion dollar computing deal, and Meta finished the first phase of its giant Prometheus data center. Alliances shifted: OpenAI models are coming to Google Cloud, and Replit made OpenAI its default. Google and Meta built tools offering alternatives to Nvidia's chips, while OpenAI detailed its own chip. Skeptics questioned a startup's soaring valuation. Meanwhile, the World Bank pushed small AI for basic phones, Qualcomm expects powerful models on phones by 2028, and Recursive released a laptop coding model. The pushback gets organized - openDemocracy reports Amazon uses an AI tool called Atlas to score warehouses on their likelihood of unionizing. Amazon says it works only at site level and doesn't monitor individuals. Shoppers are uneasy about AI-driven pricing, and the FTC has warned about personalized prices. Smart TV owners are using AI tools to strip out ads, a Samsung TV researcher says OpenAI suspended their account, and gamers want Steam to fil...

Support The Automated Daily directly: 15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going: theautomateddaily.com/support Please support this podcast by checking out our sponsors: - Effortless AI design for presentations, websites, and more with Gamma - https://try.gamma.app/tad - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad - Lindy is your ultimate AI assistant that proactively manages your inbox - https://try.lindy.ai/tad Today's topics: Artemis Three test plan - NASA detailed a complex Artemis Three mission in which astronauts would test lunar-lander designs from Blue Origin and SpaceX while remaining in Earth orbit. The demonstrations are intended to reduce risk before a planned crewed lunar landing in 2028. [34][41] Four launches in one day - Four orbital launches took place on October 23, including two Chinese remote-sensing missions, a Rocket Lab flight carrying radio-monitoring satellites, and a SpaceX Starlink launch. The busy day illustrates the increasingly routine pace of global launch activity. [210][212] Giant black holes and dark matter - Researchers found that hypothetical black holes heavier than entire galaxies cannot account for a significant share of dark matter. Their distribution would have left a grainy signature in the cosmic microwave background, but that signature is absent. [46][52] Mars canyon remains active - Scientists traced a marsquake to a deep fault beneath Valles Marineris, showing that the solar system’s largest canyon remains tectonically active. The finding changes the picture of Mars as an entirely quiet geological world. [76][82] Asia joins the LaserSETI network - Two LaserSETI detectors in Saudi Arabia are now scanning the sky for brief optical signals, creating the network’s first observing site in Asia. The location expands coverage for searches involving technosignatures and unusual astronomical flashes. [91][92] White dwarfs reveal unexpected colors - Models of human color vision suggest that some white dwarfs would appear green, pink, purple, yellow or cyan up close. Their unusual colors reveal the elements and planetary debris present in these dense stellar remnants. [106][109] Episode Transcript Artemis Three test plan We begin with NASA’s evolving route back to the Moon. The agency has presented new details for Artemis Three, a roughly two-week mission planned for 2027. Rather than landing astronauts on the lunar surface, the crew would remain close to Earth and practice working with test versions of two competing Moon landers. The plan calls for NASA’s Orion spacecraft to meet and dock separately with vehicles developed by Blue Origin and SpaceX. Astronauts would evaluate how people enter, occupy and operate these systems before either design carries a crew near the Moon. Three major rockets would be involved, making this an unusually demanding launch campaign even though the activity would happen in Earth orbit. The significance is risk reduction. NASA wants to discover integration problems close to home rather than during a lunar mission, while also keeping both lander programs moving toward a possible surface return in 2028. The schedule remains ambitious, and an exact launch date has not been fixed, but the agency now has a clearer rehearsal plan for one of the hardest parts of Artemis. [32][34][41] Four launches in one day It was also a crowded Friday for launch teams around the world, with four missions reaching orbit on October 23rd. Rocket Lab launched three HawkEye 360 satellites from New Zealand. Those spacecraft are designed to locate and study radio transmissions from orbit, with applications ranging from maritime tracking to detecting navigation-signal interference. China conducted two...

Support The Automated Daily directly: 15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going: theautomateddaily.com/support Please support this podcast by checking out our sponsors: - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad - Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron - Prezi: Create AI presentations fast - https://try.prezi.com/automated_daily Today's topics: Microsoft AI Beats Doctors at Diagnosis - Microsoft's MAI-DxO medical AI scored 83% on complex NEJM diagnosis cases versus about 36% for physicians, per a study in Science, and ordered fewer tests. Hospital trials in Boston are next. OpenAI Nears $1.25 Trillion Valuation - OpenAI is close to raising up to $30 billion at a $1.25 trillion pre-money valuation, led by UAE investors and Coatue, as a bridge ahead of a possible IPO, with possible UAE compute capacity included. Flapping Airplanes Valuation Hype Questioned - AI startup Flapping Airplanes may raise at a $5 billion valuation, up from $1.5 billion, but critics say venture capital hype outpaces evidence on sample efficiency and GPU efficiency claims. OpenAI Jalapeño Custom Inference Chip - OpenAI hardware VP Richard Ho explains Jalapeño, a custom AI inference chip pairing HBM memory with dedicated cores for better efficiency, lower latency and performance per watt. Microsoft Decision-1 Compact Model - Microsoft released Decision-1, a 9B-parameter model built on Qwen3.5 for fast routing, classification and verification, available only via Microsoft Foundry and OpenRouter APIs. Gemini 3.5 Pro Never Shipped - Sundar Pichai says Google prioritized agentic and coding gains in Gemini Flash models, and by the time those improvements were ready, Pro pretraining was done, explaining the missing Gemini 3.5 Pro. Perplexity JOLT Teacher-Student Training - Perplexity Research's JOLT jointly trains teacher and student models in self-distillation, improving reasoning, coding and agent benchmarks like MATH, LiveCodeBench and Terminal-Bench. NVIDIA JCA Video and 3D Motion - NVIDIA Research's Joints, Camera, Action! generates video together with synchronized, replayable 3D rig motion from text, supporting many skeletons and video-to-motion conversion. WorldTUI Neural Terminal Simulator - WorldTUI is a 360M-parameter model that imitates a terminal with no real shell, predicting screen changes per keypress and running in the browser, a step toward learned interactive interfaces. Enron Emails Become AI Knowledge Base - The Enron Brain project turned 45,000 Enron emails into a source-backed, time-aware AI knowledge base, with an audit finding 93% of sampled facts fully supported. Claude Outputs Carry Invisible Watermarks - Anthropic's latest Claude models embed invisible statistical text watermarks across the API, Claude Code and Desktop, driven by EU AI Act transparency rules, to flag AI-generated content. METR Red-Teams OpenAI Monitoring - A METR and Redwood Research red-team exercise found vulnerabilities in OpenAI's internal misalignment monitoring for AI agents, which OpenAI says it has since patched. Anthropic Donates Agent Skills Standard - Anthropic donated the Agent Skills open standard, used by 70-plus platforms including ChatGPT and GitHub Copilot, to the Linux Foundation's Agentic AI Foundation. Compute Treaty to Slow AI - An essay argues a unilateral US AI slowdown fails, proposing a verifiable US-China treaty built on compute governance, chip licensing and training cluster limits. Samsung TV Hacker Suspended by OpenAI - In the Samsung smart TV privacy story, the researcher who found on-device AI sound listening features says OpenAI suspended their account after the investigation. - AI Startup Rumored to Raise at $...

Support The Automated Daily directly: 15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going: theautomateddaily.com/support Please support this podcast by checking out our sponsors: - Consensus: AI for Research. Get a free month - https://get.consensus.app/automated_daily - Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron - KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad Today's topics: Samsung TV AI Listens for Dogs and Babies - A hacker who reverse-engineered a new Samsung smart TV found its onboard AI listening for dog and baby sounds, and his OpenAI account was then suspended. The case fuels debate over smart TV privacy, ambient audio collection, and the right to inspect hardware you own. iPhone Hacked Live at Hexacon - Security researchers showed iPhone exploits at Hexacon in Paris, including a one-click attack on Signal and iMessage-related scenarios. It is a notable mobile security result against Apple's iOS and encrypted messaging apps. Canada's Bill C-22 Surveillance Fight - The EFF warns that Canada's Bill C-22, the Lawful Access Act, could force encryption backdoors and year-long metadata retention, with data sharing to foreign police. The bill passed the House of Commons and is now before the Senate. Kimi K3.1 Pro Open-Weight Model - Kimi released K3.1 Pro, its strongest open-weight AI model, with a 1-million-token context window and gains in coding, agentic tasks, 3D creation and finance. It is available via API now, with weights to follow, and was trained with cyber-safety limits. Block's Dessa Speech-Only Voice AI - Block unveiled Dessa, a full-duplex speech model trained only on human conversation, not text, that can listen and talk at once and run locally on a MacBook Pro. It aims to capture emotion, laughter and interruptions that text-based voice AI loses. Twelve Years of Distributed SQLite - rqlite creator Philip O'Toole reviews twelve years of distributed SQLite, from Raft replication to Litestream, dqlite, cr-sqlite and libSQL/Turso. He argues each tool answers a different need and that commercial money went to managed cloud platforms. OBS Studio 33 Major Release - OBS Studio 33.0.0 updates the browser-source engine, overhauls plugin loading, adds NVENC on Arm64 and fixes many crashes across Windows, macOS and Linux. Developers get new APIs and higher toolchain requirements. Kevin Bacon Isn't Hollywood's Center - An interactive analysis of Six Degrees of Kevin Bacon finds Samuel L. Jackson is the true center of Hollywood, followed by Willem Dafoe and Robert De Niro. A simulation picks Chloë Grace Moretz as the likeliest Gen Z successor. - Local Wikipedia Brings Nearby Wikipedia Articles to Your Area - Kimi Launches K3.1 Pro Open-Weight Model - Researchers Claim iPhone Hack Demonstrations at Hexacon - EFF Warns Canada’s C-22 Could Expand Surveillance Powers - Block Unveils Dessa, a Speech-Only Conversational AI - The Real Kevin Bacon Reimagines Hollywood’s Center - Twelve Years of Distributed SQLite - Hacker Finds Samsung TV AI Listening for Dog and Baby Sounds - OBS Studio 33.0.0 Adds Major Browser, Plugin, and Stability Upgrades Episode Transcript Samsung TV AI Listens for Dogs and Babies Let's start with that TV. A report shared on Reddit describes a hacker who reverse-engineered a new Samsung television and found its built-in AI analyzing household audio for sounds linked to dogs and babies. That's not something most owners would expect a TV to be doing in the background. Adding to the strangeness, his investigation reportedly ended with his OpenAI account being suspended. The discussion quickly turned to a familiar question: if you buy a device, shouldn't you be allowed to look inside it and see what it's doing? It's...

Support The Automated Daily directly: 15 free episodes a day, 5 editions, 3 languages, about $4 a day to produce. A coffee keeps it going: theautomateddaily.com/support Please support this podcast by checking out our sponsors: - Prezi: Create AI presentations fast - https://try.prezi.com/automated_daily - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad - Consensus: AI for Research. Get a free month - https://get.consensus.app/automated_daily Today's topics: China-linked AI-powered hacking campaign - Britain's National Cyber Security Centre is investigating a China-linked hacking campaign that targeted hundreds of organizations. Microsoft says the attackers used AI and sometimes breached cloud infrastructure in under three hours. AI agents and PHP vulnerability attacks - In an update, a PHP vulnerability reportedly found by AI agents has been targeted in millions of hacking attempts against websites. The case adds pressure on governments and companies to strengthen AI and application-security defenses. Moonshot limits Kimi cyber capabilities - Chinese AI startup Moonshot removed cyber-related training data from its Kimi K3.1 Pro model to reduce offensive hacking abilities. Its internal cyber benchmark score fell from 90.5% to 81.0% while general coding skills were retained. Microsoft exits UK Nscale lease - Microsoft withdrew from a UK data-center lease with Nscale after delays in securing electricity grid access. The move highlights power constraints and timing risks in AI infrastructure projects. Neocloud debt pressures mount - Investors are reassessing neocloud companies that build AI data centers and rent out GPUs. Rising borrowing costs, debt levels, power access and project delays are squeezing margins. SK Hynix shelves Solidigm IPO - SK Hynix abandoned plans for a US IPO of its flash-memory subsidiary Solidigm. The decision reflects investor caution about AI capital spending despite strong demand for AI hardware. OpenAI valuation and Anthropic IPO - OpenAI is reportedly heading toward a $1.25 trillion valuation, while a potential Anthropic IPO could deliver big gains for early investors. The news fuels debate over an AI valuation bubble. Meta acquires AI analytics startup Sundial - Meta acquired Sundial, an AI-powered data analytics platform founded by two former Meta employees. The deal reflects Big Tech's ongoing race for specialized AI talent and products. Private 4G and 5G military networks - Brazil's Navy tested a private 4G network combining LTE, mission-critical radio and LEO satellite links with Motorola Solutions. Teltronic deployed a rapid private 5G network for Spain's Atlas26 military exercise. Episode Transcript China-linked AI-powered hacking campaign Britain's National Cyber Security Centre is investigating a China-linked hacking campaign said to have hit hundreds of organizations around the world. According to Microsoft, the attackers leaned on AI to run their operations, and in some cases went from first contact to access inside cloud infrastructure in under three hours. That speed is the real story here. Intrusions like this used to take days or weeks of careful manual work. If AI can compress that into an afternoon, defenders have far less time to notice and react, and sophisticated attacks become cheaper for anyone who wants to try them. AI agents and PHP vulnerability attacks Staying on that theme, there's an update on the PHP story we've been tracking. The vulnerability that AI agents reportedly uncovered has now been targeted in millions of hacking attempts against websites worldwide. It's a clear example of the full cycle speeding up, from discovering a flaw to exploiting it at scale, and it's adding to the pressure on governments and companies to tighten their appli...