Daily brief

Saturday, 3 October 2026

The last 24 hours' hottest news in AI, sorted by field, each story with a line on why it matters.

Story of the day

The single most consequential story of the day.

ElevenLabs valuation
$22B
audio needed to clone a voice
10 sec
local enterprise customers
70+
ANZ headcount since April
3→22

ElevenLabs pitches brands on cloned voices from 10 seconds of audio

ElevenLabs' ANZ chief says marketers no longer need to care whether a voice is real, as the $22B startup pushes its new v4 model and voice marketplace into Australian and New Zealand advertising. The company's local headcount has grown from three to 22 since April, with over 70 enterprise customers signed up.

Why it's the story of the dayIt shows how fast AI voice cloning is becoming normal in everyday advertising and customer service.

Startup of the day

A young company people are talking about today.

Modal

Modal launches VM Sandboxes so AI agents get a full computer

Modal made VM Sandboxes generally available, giving AI agents a full Linux virtual machine instead of a stripped-down container, with sub-second start times. Early users have already launched more than 20 million of these VMs, including Linear's coding agents.

Why it's worth a lookAs agents take on bigger jobs, like running whole dev environments, they need more than a lightweight sandbox, and Modal is racing to supply that.

Modal ↗

Where the industry gathered

The conferences, keynotes and festivals where people met in person.

1 events

AI Film Summit Los Angeles 2026

Conference24 Oct
Los Angeles, City Club Los Angeles

It hosts the Tyrannus Angel Awards recognising AI-made films.

  • 34 films selected from 636 submissions across 84 countries
  • Winners to be announced at the event on 24 October
The Manila Times ↗

Models & labs

New models, the labs behind them, and the rules they answer to.

Funding

Anthropic to invest $100 million to train AI engineers

Anthropic is funding the Claude Frontier Academy to train 10,000 'frontier deployed engineers' by the end of 2027, starting with staff from Accenture, Morgan Stanley and Novo Nordisk. The move comes as Anthropic heads toward a possible $2 trillion IPO.

Why it mattersIt shows Anthropic trying to fix a shortage of people who know how to actually deploy AI inside businesses.

CNBC ↗
Safety

OpenAI fires three staff over leaks to an AI safety group

OpenAI said it 'parted ways' with three members of its internal safety team for mishandling sensitive information shared with an outside AI safety group. The move follows a cancelled model release and reports of its agents breaching outside websites.

Why it mattersIt raises questions about how openly AI labs will work with independent safety watchdogs.

New York Post ↗
Funding

Nvidia and SoftBank complete $20B investment in OpenAI

Reports say Nvidia and SoftBank finalised a combined $20 billion investment in OpenAI, underscoring how scarce frontier AI computing power has become.

Why it mattersIt is one of the largest single confirmations yet of money flowing into OpenAI's compute buildout.

Moomoo ↗
Model

Cloudflare releases Clef open-weight decision models

Cloudflare launched Clef and Clef-flash, open-weight decision models built on Qwen backbones that can handle yes/no questions, multiple choice and rankings. They also handle images and video, support a 64k context window, and are available on Hugging Face.

Why it mattersOpen-weight models that beat Jev on accuracy benchmarks give developers a free alternative to proprietary decision systems.

The Register ↗

Agents & coding

AI that acts on its own, and the tools developers build with.

Product

Cline open-sources an SDK for building physical 'Muse' gadgets

Cline released an open-source SDK letting hobbyists turn an ESP32 board or Raspberry Pi into a device that connects to its Muse assistant, handling sysadmin tasks or smart-home control.

Why it mattersIt lets independent developers build their own hardware around an AI agent, not just software.

GitHub ↗
Product

AWS launches Well-Architected Agent to audit cloud environments

AWS announced a public preview of Well-Architected Agent, an AI service that analyzes cloud environments and recommends improvements to cost, security, performance and resilience. It covers 65+ AWS services in three US regions.

Why it mattersAn agent that plays the role of an expert architect could help customers move faster through cloud optimization without hiring consultants.

PYMNTS.com ↗
Research

Louisiana's AI data centre boom raises questions over power and water

Meta is expanding its Richland Parish data centre to 5 gigawatts, while Amazon and Anthropic build competing facilities across the state. The scale raises questions about power supply, water use and environmental impact.

Why it mattersThe shift of AI infrastructure to the South is reshaping state economies and putting pressure on utilities and water supplies.

Tulane Law School ↗
Product

Perplexity Computer integrates TradingView charts for finance tasks

Perplexity's agent platform now displays financial data in TradingView's chart formats. Users can ask agents to show price history, add overlays and compare stocks, with the agent building custom views for each request.

Why it mattersAgents that can show data in formats users already know make complex analysis easier to follow.

TradingView ↗
Product

NetApp and Iterate pair private AI with enterprise storage

NetApp's AIPod Mini appliance pairs its storage with Iterate.ai's Generate software, shipping with over 200 agent templates and 800 tools. In one insurance case, accident-report analysis dropped from eight hours to eight minutes; in healthcare, it found $17.4 million in denied claims for a $150M hospital.

Why it mattersTurnkey private AI systems let enterprises run agents on their own data without sending it to a third party.

SiliconANGLE ↗
Research

McKinsey: agents still cost more than they save, for now

Only 37% of companies see positive earnings impact from their AI programs, and agent workloads cost five to 30 times more compute than chatbots. McKinsey says agents are maturing, with frameworks like LangGraph adding the oversight needed for longer workflows.

Why it mattersBusinesses are spending heavily on agents, but returns are lagging; success depends on the technology and practice maturing together.

cio.com ↗
Research

Bain: most companies lack the systems to make AI agents work

Bain's Technology Report 2026 says the bottleneck in AI-assisted software development has shifted from model capability to the engineering systems around agents. Most companies have the AI tools but few have built the architecture to make them effective, and review time has risen 91% even as developers complete 21% more tasks.

Why it mattersAs frontier models improve, the gap between what they can do and what companies can actually deploy grows wider, pointing to where the real work now is.

Bain ↗
Research

Open-weight models now majority of tokens on Vercel gateway

Open-weight AI models handled 56% of token volume on Vercel's AI Gateway in August, up from 13% in April, marking the first month they made up the majority. However, they represented only 14% of spending, as higher-priced closed models continued to dominate revenue.

Why it mattersFree models are scaling fast enough to become the norm for developers, even as a few proprietary models still capture most of the money.

IT Brief New Zealand ↗

Video generation

The models and tools that make video.

Model

Reka releases Inverse Dynamics Model for robot training

Reka published its Reka Inverse Dynamics Model (RIDM), which extracts low-level motor commands from video to help train interactive world models for robots. Trained on video games, it transfers to real-world video and is available on Hugging Face under Apache 2.0.

Why it mattersBetter simulation training could speed up robot learning without expensive real-world trials.

reka.ai ↗
Model

Nvidia releases PixelUMM for image and video generation

Nvidia published PixelUMM, a 15.2-billion-parameter encoder-free model that can generate and understand text, images and video directly in pixel space. It supports text-to-image generation, image-conditioned text, video understanding and video generation.

Why it mattersA unified model that works across text, images and video in one architecture could simplify multimodal AI pipelines.

Hugging Face ↗
Model

Black Forest Labs releases FLUX 3 Image with 4K generation

Black Forest Labs launched FLUX 3 Image, which generates images up to 4K resolution, positions subjects with bounding boxes, edits specific areas without changing other pixels, and accepts up to 10 reference images. An open-source version is coming soon.

Why it mattersPrecise control over image composition and the promise of an open version could make 4K image generation accessible to smaller creators.

GIGAZINE ↗
Model

Tavus' Griffin model passes for human on live video calls

Tavus previewed Griffin, a 'Human Interaction Model' that renders a lifelike person who watches, listens and reacts live. In testing, 48% of users thought Griffin-Lite was human, up from 2.4% before.

Why it mattersAI video that fools nearly half of people in real-time calls marks a sharp step toward synthetic avatars that could replace human presence in some contexts.

The Rundown AI ↗
Research

AI video tools reshape content creation, raising verification challenges

Virtual presenters made with AI now appear in politics, news, cooking, education and entertainment. Thousands of professional-looking clips circulate without clear attribution of who made them.

Why it mattersAs AI video floods the web, viewers will struggle to know what they are watching, making verification skills as important as the technology itself.

dailybeirut.com ↗
Funding

Tripo AI wins SIGGRAPH Best in Show, raises $350M

Tripo AI, a Beijing-founded generative 3D tool, won Best in Show at SIGGRAPH 2026's Real-Time Live! competition and raised $350 million at a $1 billion valuation. Its SmartMesh tech generates game-ready 3D assets with texture in under two seconds.

Why it mattersWinning SIGGRAPH's top prize and raising major funding shows AI 3D generation is becoming essential for creators.

Animation World Network ↗
Policy

Arizona court tosses sentence over AI video of dead victim

An Arizona appeals court vacated Gabriel Horcasitas's 10½-year manslaughter sentence after finding the AI-generated video of his victim shown during sentencing was 'not reliable' and a 'fundamental error'. The video showed an AI recreation of Christopher Pelkey forgiving his killer, created by Pelkey's family and presented as a victim impact statement.

Why it mattersIt is the first court to rule that an AI-generated victim impact video was too prejudicial to use in sentencing, setting a legal boundary for AI in the courtroom.

New York Post ↗
Product

ByteDance's Doubao readies personal AI agent 'Spell'

ByteDance's AI assistant Doubao is preparing a personal AI agent codenamed Spell, which has been in internal testing since April. The project moves Doubao from answering questions toward completing tasks on behalf of users, and a public launch is expected soon.

Why it mattersMajor AI labs worldwide are racing to ship personal agents that act autonomously, and ByteDance is catching up to Meta and OpenAI.

TechNode ↗

Creator tools

The apps and platforms people make video with.

Product

ElevenLabs pitches brands on cloned voices from 10 seconds of audio

ElevenLabs' ANZ chief Damian Naughton says it no longer matters if marketers use synthetic voices, as its v4 model nears human-level realism. The company's local headcount grew from three to 22 and it now has over 70 enterprise customers.

Why it mattersIt shows how fast AI voice cloning is becoming normal in everyday advertising and customer service.

Mumbrella ↗
Product

Snowflake data gets wired directly into marketing campaigns

A new integration lets marketers turn customer data and AI models already stored in Snowflake into audiences and campaigns without a new request to the data team each time. It connects Snowflake's Cortex and Customer 360 tools to Insider's engagement platform.

Why it mattersIt cuts the time between having customer data and actually using it to reach customers.

Medium (Snowflake) ↗

Media & entertainment

AI in the films, shows, micro-dramas, music and ads people watch, the studios and platforms making them, and the rules for AI-made content.

Product

AI music maker Suno adds spoken-word generation

Suno's new Speech feature, in public beta, generates voiceovers and background music together from a script or a short prompt. Clips can run up to about eight minutes, and the music can be toggled off for plain speech.

Why it mattersIt pushes Suno beyond music into the same AI-voice market as ElevenLabs and Adobe.

The Verge ↗
Product

Mastercard uses a K-drama microdrama to teach payment safety

Mastercard and McCann Singapore launched 'Tap Into My Heart', a three-episode K-drama style microdrama about fraud and payment security, running across six Southeast Asian countries over eight weeks.

Why it mattersBrands are turning to the fast-growing microdrama format to make dry safety messages watchable.

marketech apac ↗
Research

India's microdrama market set to double to ₹4,600 crore

India's microdrama sector generated ₹2,300 crore in FY26 and is projected to double to ₹4,600 crore in FY27, with 40 million viewers already paying for the short-episode format.

Why it mattersIt shows how fast short, serialised video is turning into a real business beyond a mobile novelty.

CNBC-TV18 ↗
Deal

Sean Parker rebuilds Stability AI around music

Sean Parker, who joined an $80 million rescue of Stability AI two years ago, is repositioning it as an AI toolmaker for music professionals. In late August, Stability raised $76 million from Sony, Warner and Universal, who licensed their catalogs for training.

Why it mattersMusic labels are now bankrolling AI music tools, betting the tools will create more value than they risk losing to piracy.

TechCrunch ↗
Product

Meta's Muse connects to Spotify for context-aware playlists

Meta linked its Muse personal agent to Spotify so users can ask it to make playlists based on context, like picking music for a road trip. Muse also works with Ticketmaster and Plaid.

Why it mattersAI agents that know what you are doing, not just what you ask, could change how people interact with media and shopping across apps.

Mashable ↗
Product

Adobe Firefly adds music, speech and sound effects to creative workflow

Adobe made Generate Music, Generate Speech and Generate Sound Effects broadly available in Firefly, letting creators build finished audio alongside visual assets. Generate Music tunes tracks to video length and mood, while the other tools handle voiceovers and sound effects.

Why it mattersTreating audio as an editable layer inside one creative tool could cut production time for video creators and agencies.

WeRSM ↗
Research

Astana AI Film Festival showcases 25 AI films from 125 countries

The Astana AI Film Festival (AAIFF 2026) ran October 1–3, receiving over 8,000 applications and selecting 25 finalists. It aims to become annual and launch a streaming platform, Mooncast.ai.

Why it mattersA dedicated festival for AI cinema signals that AI films are moving beyond novelty to a form worthy of curated exhibition.

The Astana Times ↗
Product

IMDA and TikTok launch Singapore microdrama training programme

Singapore's IMDA and TikTok launched a six- to nine-month programme for media professionals from 20 accredited companies to learn microdrama production. Participants choose live-action or AI-enabled animation tracks and produce 50–100 episodes of one to three minutes each.

Why it mattersA government-backed training programme signals microdramas are becoming a mainstream media format that needs formal skill development.

CNA Lifestyle ↗
Product

Room418 adapts 'Night of the Living Dead' as vertical microdrama

L.A.-based Room418, a newly-established microdrama studio focused on horror, is adapting the classic 'Night of the Living Dead' for its vertical video app. The platform will launch with acquired shows and original productions.

Why it mattersHorror as a dedicated vertical format shows genres are carving out their own microdrama niches.

Deadline ↗
Research

Microdrama platform founders say Hollywood is now paying attention

At TheGrill 2026, founders of Shortical, Mansa and aTwist said Hollywood is knocking on their doors as traditional pipelines shrink. Shortical is on track for $100 million in revenue by year-end.

Why it mattersTraditional entertainment leaders are now backing vertical formats, signalling a permanent shift in how stories reach audiences.

TheWrap ↗

Chips & compute

Chips, data centres and the power behind them.

Product

AMD pitches Ryzen chips as the engine behind agentic AI on PCs

AMD says its 'Zen 5' Ryzen processors are built to handle the local tasks AI agents need, like opening files, running code and coordinating multiple sub-agents, rather than just chatbot responses.

Why it mattersIt shows chipmakers now designing hardware specifically for AI that takes actions, not just answers questions.

AMD ↗
Deal

Amazon to offload $8B in AI chips via special vehicle, lease them back

Amazon plans to move about $8 billion in Grace Blackwell chips to a special-purpose vehicle, which will issue debt and a 10% equity stake to investors. Amazon will then lease the chips back, moving both the chips and debt off its books.

Why it mattersTech giants are restructuring AI debt to improve their credit ratings and hide the scale of their spending.

SiliconANGLE ↗
Research

$6 trillion in annual AI revenue needed to justify data-centre spending

Bain estimates hyperscaler capex could hit $780 billion in 2026 and $1.5 trillion by 2031. At 25% of revenue, that implies AI needs $6 trillion in annual revenue, but consumer and enterprise markets may only reach $1.8 trillion.

Why it mattersThe gap between spending and revenue is widening, forcing the industry to find entirely new markets or face losses.

networkworld.com ↗
Research

CoreWeave backlog tops $104 billion as demand for AI compute surges

CoreWeave's $104 billion revenue backlog includes about 40% expected in the next two years, 39% in two to four years and 21% beyond that. Q2 revenue was $2.575 billion, up 112%, with GAAP EPS of -$1.14. A 25% price increase across services took effect in July, and managed inference annual recurring revenue grew from $1 million to over $100 million.

Why it mattersCoreWeave's backlog and pricing power show how desperate companies are for AI compute, even as the company burns cash to scale.

The Globe and Mail ↗
Deal

Moonshot AI trains on 20,000 Nvidia chips via Alibaba partnership

Beijing-based Moonshot AI powers its Kimi models using about 20,000 Nvidia chips leased from backer Alibaba. Reports suggest the cluster uses H200 chips, though Alibaba disputed that.

Why it mattersChina's top AI labs remain dependent on Nvidia's hardware despite US export controls and Beijing's push for alternatives.

Stocktwits ↗
Product

Apple's Mac Studio M5 Ultra brings frontier AI models to the desktop

Apple released the Mac Studio with M5 Ultra, with up to 256 GB of unified memory. The machine can run large language models, agentic AI frameworks and high-end video generation on-device. The $11,000 top configuration is designed to bring frontier-level AI compute to a desk without cloud calls.

Why it mattersOn-device AI is becoming viable for professionals, shifting where AI work happens and who controls the data.

WIRED ↗
Research

Cerebras reports 5x inference throughput gain from disaggregation

Cerebras Systems said disaggregation, running prefill and decode phases on separate hardware pools, lifted inference throughput 5x. The gain came with the same number of systems and no loss in token speed.

Why it mattersSplitting inference work between different hardware could let AI systems serve more users without slowing down each one.

Unite.AI ↗
Research

CoreWeave's backlog and growth outpace the market, but debt looms

CoreWeave trades at $85.93 with a 24/7 Wall St. price target of $163.81 (91% upside). The inference provider has a $104 billion signed backlog and revenue growing over 100% annually, but faces mounting losses, heavy debt and negative free cash flow.

Why it mattersCoreWeave's finances show the paradox of the AI boom: explosive growth and massive pre-orders alongside unsustainable losses.

24/7 Wall St. ↗

Policy & safety

Laws, lawsuits and safety incidents.

Policy

Court blocks Minnesota's AI 'nudification' ban after xAI lawsuit

A US federal appeals court halted Minnesota's first-in-the-nation ban on AI-generated fake nude images, siding with xAI, which argues the law is unconstitutional. A lower court had earlier rejected xAI's request to stop the law.

Why it mattersIt is a test of whether states can restrict AI-made explicit imagery at all.

WTVB / Reuters ↗
Policy

Prosecutors seek 4 years for $8M AI music streaming fraud scheme

A man who used AI-generated music and billions of fake streams to collect $8 million in royalties pleaded guilty in March. He is now seeking probation, arguing that losses were spread so widely that no individual artist suffered meaningful harm.

Why it mattersCourts will now weigh whether diffuse fraud across millions of artists is less serious than targeted theft.

lawcommentary.com ↗
Policy

Kelly Clarkson says AI music should have its own separate chart

The singer and talk show host said she supports AI in music but believes AI-created music should be separated from human-made music on charts. Clarkson compared mixing AI and human artists to having robots compete with humans in the Olympics.

Why it mattersAs AI music floods streaming platforms, artists and listeners are pushing for clearer boundaries between human and machine creation.

Digital Music News ↗
Policy

UK personality rights law exposed by AI voice cloning, 80+ figures demand protection

About 80 actors, performers and public figures backed the 'Save Our Voices Now' campaign launched 28 August, calling for stronger UK legal protection against unauthorized voice cloning by AI. The campaign highlights gaps in UK law on personality rights.

Why it mattersVoice cloning is outpacing the law; countries without strong personality-rights protections are racing to close the gap.

Law360 ↗
Policy

Amazon commits $1B to data centre communities for training, energy and water

Amazon announced Built Together, a $1 billion five-year community investment programme for data centre areas. It funds free education for 300,000 students, upgrades for 30,000+ homes and schools, and 8 billion gallons of annual water replenishment.

Why it mattersAs hyperscalers face data-centre moratoriums in over 100 US communities, they are offering public investment to reduce local resistance.

Unite.AI ↗
Safety

J&K Police warn of AI voice-cloning scams

Jammu and Kashmir Police issued an advisory about cyber criminals using AI to imitate voices of relatives, employers and officials in fraudulent calls. Police advised people to verify callers independently and not share personal information during unsolicited calls.

Why it mattersVoice cloning scams are spreading fast enough that police are now warning the public about them.

The News Now ↗
Safety

Google says AI is accelerating vulnerability discovery

Google's Threat Intelligence Group found that disclosed vulnerabilities doubled in 2026 and high-risk flaws increased significantly. The report warns that attackers are moving faster to exploit newly disclosed vulnerabilities, creating challenges for security teams.

Why it mattersAI is making both finding and exploiting security holes faster, forcing companies to rethink how they patch systems.

Petri IT Knowledgebase ↗
Policy

Maryland governor pushes state AI regulation as federal efforts stall

Maryland Gov. Wes Moore, chair of the National Governors Association, is leading a bipartisan effort among state leaders to develop an AI regulatory framework. He criticised President Trump's White House meeting with AI executives as a 'billionaire boys club'.

Why it mattersWith Congress stalled on AI rules, governors are stepping in, but states have limited authority over technology that crosses borders.

PYMNTS.com ↗
Policy

Court orders resentencing after AI victim statement swayed judge

An Arizona appeals court ordered a man convicted of manslaughter to be resentenced after finding that an AI-generated video of the victim fundamentally affected the judge's sentencing. The judge had called the synthetic video 'genuine', but the court ruled it 'not reliable' and prejudicial.

Why it mattersCourts are grappling with whether AI-made evidence should be allowed, after it appears to have changed a real sentence.

PetaPixel ↗
Policy

UN warns that AI will shape the world's future alongside climate and conflict

UN Secretary-General Antonio Guterres, in his final address before stepping down, said artificial intelligence alongside climate change, wars and inequality are the four tests that will shape the world's future.

Why it mattersThe UN's top official put AI at the centre of global governance, elevating it to the same level as existential threats like climate and conflict.

Al Jazeera ↗

Robotics

Robots, self-driving and AI in the physical world.

Policy

Waymo robotaxis approved to expand in California

The California Public Utilities Commission approved Waymo's robotaxi expansion beyond city centres into neighbouring counties, including the heart of Silicon Valley. The approval lets driverless taxis operate further across Los Angeles and the San Francisco Bay Area.

Why it mattersWaymo can now reach more customers as one of the few robotaxi services still operating after Cruise's suspension.

Mashable ↗
Product

Tesla delivers 486,000 vehicles in Q3, beating estimates

Tesla delivered over 486,000 vehicles in the third quarter, beating analyst consensus of about 462,000. The result was a 2% year-on-year drop but exceeded expectations given that government EV tax breaks expired in September 2025.

Why it mattersStrong delivery numbers keep investor focus on Tesla, though the real test will be how AI and robotics pay off.

The Globe and Mail ↗
Funding

FieldAI raises $700M at $10B valuation for robot navigation

FieldAI, which builds AI models that let robots navigate without maps or GPS, is raising $700 million in funding at a reported $10 billion valuation. That is five times what the company was valued at a year earlier.

Why it mattersFieldAI's funding jump shows investor appetite for AI that cuts the cost of setting up autonomous robots.

SiliconANGLE ↗
Product

Faraday Future shows its EAI robotics platform at IROS 2026

Faraday Future showcased its EAI Robotics World 2.0 platform at the IEEE/RSJ International Conference on Intelligent Robots and Systems in Pittsburgh. The company demonstrated robot products, introduced a 'Built in USA' acceleration programme, and recruited development partners.

Why it mattersFaraday Future is moving from cars to a full AI robotics platform with industry applications.

Voice of Alexandria ↗
Deal

Eni and Generative Bionics partner on humanoid robots for industry

Italian energy company Eni and Generative Bionics signed a memorandum of understanding to explore industrial uses for Generative Bionics' GENE.01 humanoid robot in inspection, teleoperation and remote assistance. The partnership also covers materials, footwear design, sensor integration and battery recycling.

Why it mattersMajor energy companies are starting to test humanoid robots in real industrial work, not just research.

AI Insider ↗
Research

Tesla reports Q3 deliveries of 486,532 vehicles and 13.7 GWh storage

Tesla delivered 486,532 vehicles in the third quarter of 2026 and deployed 13.7 GWh of energy storage products. Model 3 and Model Y vehicles accounted for 478,237 deliveries.

Why it mattersTesla's delivery numbers are watched as a proxy for EV demand and Elon Musk's ability to execute on his AI and robotics ambitions.

Unite.AI ↗
Research

Tesla's sell ratings hit lowest since April 2023 as Musk chases AI

Sell ratings on Tesla account for just 13.1% of analyst recommendations, the lowest share since April 2023, as Wall Street increasingly hesitates to bet against Musk's pivot toward autonomous vehicles and humanoid robots. Hold-equivalent ratings have risen to their highest level in over two years.

Why it mattersAnalysts are rewarding Musk's long-term vision in robotics and AI even as Tesla's car business faces near-term headwinds.

The Edge Malaysia ↗
Product

Pi now supports MCP

Physical Intelligence added MCP (Model Context Protocol) to Pi's core after previously dismissing it. The company says MCP has improved over the past year, and the changes needed to integrate it also make it easier to use other tools like Jev within Pi.

Why it mattersMCP is becoming the standard way AI agents talk to tools, so Pi backing it signals the protocol is winning.

Ben's Bites ↗

Science & health

AI in biology, medicine, climate and maths.

Research

Google introduces SynthID Bio watermarking for AI-designed proteins

Google DeepMind released SynthID Bio, a technology that embeds an imperceptible, verifiable watermark into AI-designed proteins while preserving their biological function. Lab tests show watermarked designs matched the performance of unwatermarked versions.

Why it mattersWatermarking AI-made biology could help detect dangerous pathogens designed by malicious actors.

blog.google ↗
Research

Colossal bets on 'promptable biology' within 10 years

Colossal, now valued at over $10 billion, set a 24-month timeline for ex utero mammals and a 10-year bet that AI will design organisms from text prompts alone. The company has already achieved 300+ genetic edits at 90% efficiency.

Why it mattersIf AI can soon design living things on demand, the implications for medicine and biosecurity are huge.

StartupHub.ai ↗
Deal

BioMarin taps Profluent's AI protein models for enzyme therapy

BioMarin Pharmaceutical entered a research collaboration with Profluent to use its AI protein language models to design novel enzyme therapy candidates. Profluent's models are trained on over 115 billion unique protein sequences.

Why it mattersAI protein design is moving from research into the drug pipelines of major pharmaceutical companies.

AllSci ↗
Research

Family medicine doctor uses AI for precision medicine and teaching

Dr. Brittanie West, a family medicine physician at WVU Medicine United Hospital Center, uses AI for genetic risk analysis and pharmacogenomics to tailor treatments to patients' genetics. She also uses AI tools that cut the time to create a teaching module from 25 hours to four hours.

Why it mattersAI is moving from labs into everyday clinical work and training, changing how medicine is taught and practised.

wvnews.com ↗
Research

Nvidia CEO says AI must cause 'pain and suffering' before it can help the climate

Jensen Huang told The Ezra Klein Show that AI's energy demands are a temporary compromise needed to eventually address climate change. He said fossil fuels will remain necessary for several more years because there is not yet enough sustainable energy to power AI, but expects a transition afterwards.

Why it mattersThe head of the chip company powering the AI boom is openly defending continued reliance on coal, oil and gas, framing environmental harm as a price worth paying.

The Cool Down ↗
Policy

Greenpeace disrupts Nvidia at Climate Week over greenwashing

Greenpeace activists interrupted a panel at New York Climate Week where Nvidia's head of sustainability was speaking, calling the company's climate promises greenwashing. Four separate disruptions occurred during the 35-minute discussion, with protesters also objecting to data centres enabling warfare.

Why it mattersNvidia's grip on the AI industry gives it outsized influence over whether the boom harms or helps the climate, and activists are now directly challenging its public image.

Greenpeace ↗
Research

AI agents in drug discovery: a sceptic's view on what they are good for

Dr Eric Ma, a senior data scientist at Moderna, describes AI agents as 'super-powered autonomous search engines' that excel at literature review and internal knowledge search but face limits: they won't question what they believe unless told, and rely on APIs that are often poorly designed.

Why it mattersAs drug companies deploy AI agents, researchers are learning where they work well and where they can lead you astray.

Drug Target Review ↗
Deal

OpenEvidence partners Anthropic and Penn Medicine to expand AI to Africa and Asia

OpenEvidence is teaming up with Anthropic to expand its AI-powered clinical decision support to physicians in 100 low- and middle-income countries, including Uganda, Angola, Sudan, Haiti and Mongolia. Anthropic is providing Claude Credits, engineers and funding for the rollout.

Why it mattersA top AI lab is putting resources behind bringing medical AI to the world's poorest regions, where access to specialist expertise is rarest.

Fierce Healthcare ↗
Research

"Don't panic": AI's future role in healthcare debated at Trinity

At a public event at Trinity College Dublin, experts discussed AI's role in medicine, surgery and drug discovery. Opinion split: some saw cautious promise, while the Head of Trinity School of Medicine said he was "very, very worried" about generative AI dumbing down doctors' skills.

Why it mattersLeading medical voices are publicly divided on whether AI helps or harms medicine.

Trinity College Dublin ↗

Quick hits

Smaller stories worth a glance.

Deals & funding

Who raised money, and who bought whom.

14 deals

Funding rounds 5

Investors put money into a company in exchange for a share of it.

Anthropic raises new funding

Raised
$100M
Funding

Anthropic valuation not disclosed

Anthropic is funding the Claude Frontier Academy to train 10,000 AI engineers by end of 2027.

· CNBC ↗

Sony backs Stability AI

Raised
$76M
Funding

Stability AI valuation not disclosed

Stability AI raised $76 million from Sony, Warner and Universal, who licensed their music catalogs for training the company's new audio models.

Other investors

WWarnerUUniversal

· TechCrunch ↗

Acquisitions 1

One company buys another outright.

Partnerships & contracts 8

A large commercial agreement, or a government contract.

OpenAI sign a deal

Value
$20B
Partnership

Multi-year agreement

Nvidia and SoftBank completed a combined $20 billion investment in OpenAI.

Other investors

SSoftBank

· Moomoo ↗

Amazon sign a deal

Value
$8B
Partnership

Multi-year agreement

Amazon will offload $8B worth of Grace Blackwell chips to a special-purpose vehicle, which will issue debt and 10% equity to external investors; Amazon will lease the chips back.

· SiliconANGLE ↗

xAI sign a deal

Value
$1.25B/month
Partnership

Multi-year agreement

Anthropic leased the full capacity of xAI's Colossus 1 cluster (220,000 GPUs) for $1.25B per month through May 2029, with a 90-day termination clause.

· Crypto Briefing ↗

On the calendar

What's coming up, in date order, and how sure we are about each.

2 things
MTWTFSSOct282930123456789101112131415161718192021222324252627282930311Nov234567891011121314151617181920212223242526272829
  1. 24OctTyrannus Angel Awards 2026 winners announcedTyrannus Foundation3weeks away
  2. 14DecAI immuno-oncology primer at ESMO congressSITC10weeks away

Stories are summarized in plain language; follow the link before quoting a figure. Story images are shown from the publishers' own sites.