In partnership with

Good morning. It’s Tuesday, November 12th.

On this day in tech history: In 2004, the University of Ljubljana released Orange, an open-source visual machine learning environment that let users connect models, preprocessing steps, and visualizations without heavy coding. First used in academia, it helped broaden access to hands-on ML experimentation and teaching, foreshadowing today’s no-code, drag-and-drop machine learning platforms

In today’s email:

• OpenAI Is Quietly Building Your Next Health Assistant

• Google Upgrades Nano Banana and Creative Canvas

• ElevenLabs brings celebrity voices to AI, launches Scribe v2

• 5 New AI Tools

• Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

OpenAI Is Quietly Building Your Next Health Assistant

OpenAI is quietly moving beyond the chatbot that made it famous. According to people familiar with the company’s plans, it is developing AI-powered health tools that could evolve into a personal medical assistant — a kind of always-on digital doctor that listens, learns, and advises.

In recent months, the company has recruited figures with deep experience at the intersection of technology and healthcare, including Nate Gross, a co-founder of Doximity, and Ashley Alexander, a former executive at Instagram. The hires suggest OpenAI is preparing a serious entry into a market that has humbled Silicon Valley before. Google, Amazon and Microsoft have each tried to merge artificial intelligence with personal health, only to run into privacy concerns, regulatory pushback, and public skepticism.

Perhaps most intriguingly, several people close to the company say Jony Ive, Apple’s former chief designer, has been consulting on early design concepts for a device tied to the project. The collaboration, still unconfirmed, has fueled speculation that OpenAI’s health ambitions extend beyond software — toward a piece of hardware that blends voice, sensors and minimalist design in Ive’s signature style.

The company is also rolling out a practical social initiative: a free year of ChatGPT Plus for U.S. servicemembers and veterans within 12 months of leaving active duty. Beyond resume building and interview prep, the program helps with benefits, housing, education, and financial planning, supported by example chats, onboarding videos, and SheerID verification.

Speaking to a podcast, Sam Altman underscored AI’s creative limits, saying it can produce technically flawless poetry but probably won’t capture human emotion. The CEO emphasized that human creativity still has unique value even as AI outputs proliferate. Meanwhile, OpenAI is snatching up compute talent too: Intel’s CTO and Chief AI Officer Sachin Katti left to help build infrastructure for artificial general intelligence, leaving Intel CEO Lip-Bu Tan to lead the company’s AI and advanced tech efforts as it scrambles to catch Nvidia and TSMC.

The company also finds itself in a legal storm, facing subpoenas targeting nonprofits critical of its for-profit shift. OpenAI claims some of the loudest critics may have received funding from Elon Musk, though those groups remain openly critical of him. Read more.

Google Upgrades Nano Banana and Creative Canvas

Google Photos and Messages get major upgrades powered by DeepMind’s Nano Banana model. Photos adds “Help me edit” for conversational text or voice-guided edits, “Create with AI” templates, and an “Ask” button for interactive queries. Users can adjust facial expressions, remove accessories, or transform images into paintings, mosaics, or illustrations while preserving likeness via private face groups. Messages adds Remix for collaborative, real-time photo transformations directly in chats with cloud and on-device AI.

Google’s Gemini platform added upgrades with Creative Canvas and Visual Layout. Creative Canvas lets users build interactive dashboards, article summaries, or mini-games right in chat. Visual Layout handles structured, presentation-style outputs. Leaks of Nano Banana 2 show it can handle complex text prompts, remaster low-res images, and interpret text on whiteboards, a task older models struggled with.

Private AI Compute brings cloud-scale Gemini AI to devices without exposing user data. Built on custom TPUs and Titanium Intelligence Enclaves (TIE), it keeps information hardware-isolated and encrypted. The platform powers advanced AI reasoning, context-aware Magic Cue suggestions, and Pixel Recorder multilingual transcription, all while maintaining strict privacy.

On security, Google flagged PROMPTFLUX, a generative AI malware that rewrites its code in real time to evade detection. Countermeasures like Big Sleep are in the works, showing AI is now as much a cybersecurity threat as it is a productivity tool. Read more.

ElevenLabs brings celebrity voices to AI, launches Scribe v2 

ElevenLabs is expanding its AI audio footprint with two major launches and high-profile celebrity partnerships. The Iconic Voice Marketplace allows brands to license AI-generated voices of historical and living figures, including Michael Caine, Matthew McConaughey, Liza Minelli, and Alan Turing. The platform connects companies directly with rights holders, ensuring consent-based, ethically sourced use, and integrates voice cloning and archival synthesis for high-fidelity audio applications across advertising, media, and storytelling. McConaughey is leveraging the technology for a Spanish-language edition of his newsletter, while Caine’s participation highlights the marketplace’s potential for premium brand engagement.

Complementing the marketplace, ElevenLabs introduced Scribe v2 Realtime, a low-latency transcription model designed for complex audio scenarios. Supporting over 90 languages, it meets SOC 2, ISO27001, PCI DSS L1, HIPAA, and GDPR standards, offers EU and India data residency, and features a zero-retention mode. Read more.

• Metadata, not messages, is the new privacy weak spot in AI 

• Meta advances AI Infrastructure with SPICE, Omnilingual ASR as LeCun prepares to leave

• SoftBank sells its entire $5.83B Nvidia stake to fuel AI expansion, stock dips

• Epoch AI: A one-year sprint to a 1-GW data center puts Elon Musk’s xAI on track to seize the AI infrastructure crown

• Montana just passed the nation’s first “Right to Compute” law, protecting access to AI and digital tools

• A new report warns AI is transforming hospitality roles faster than employers can upskill their workforce

• Time just turned its 102-year archive into an AI you can chat with

• LinkedIn will now label cryptographically signed media with C2PA credentials

• An AI-generated country song just hit No. 1 on a Billboard chart

• The EU may bend GDPR to boost AI, setting up a fresh privacy fight

• Robinhood is letting retail traders bet on AI start-ups through a high-risk private fund

• An Indian wealth startup is swapping pricey private bankers for AI-trained college grads

• Copyleaks’ new tool claims it can tell where AI touched an image

• Big AI spending, not automation, is quietly driving layoffs and squeezing new grads

• Morgan Stanley predicts Apple could pull in $133 billion a year from humanoid robots by 2040

• Watch: Students built a robot that can withstand more than 1,000 degrees Fahrenheit

• Amazon tests robot-powered Whole Foods hybrid in Philadelphia

• Oxford engineers build robots that move and sync without brains, electronics, or motors

• Qodo’s AI becomes Monday.com’s newest “developer,” catching bugs and enforcing standards at scale

• Southeast Asia’s biggest bank DBS launches Gen AI assistant amid broader AI integration across its operations

• New study shows AI memorizes arithmetic in one brain while logic lives in another

• Moonshot AI launches Kosong, an LLM abstraction layer that powers the Kimi CLI and simplifies agent development

• Nearly a third of companies plan to replace HR roles with AI in 2026

• The Sinister Curve: AI safety updates are reshaping interactions, leaving users feeling managed and epistemically undermined

• Watch: Convertible robot changes form to take on any environment 

• To have machines make math proofs, turn them into a puzzle

• American AI Exports Program defining what “American-made AI” actually means

• Fei‑Fei Li says the next AI breakthrough won’t be in language, but in teaching machines to understand space

• JUPITER supercomputer nails full 50-qubit quantum simulation record

• Only 9% of developers trust AI code without oversight as senior engineers shift from coding to system strategy

• Watch: YouTuber builds talking robot head that answers like Greek philosopher Aristotle

• According to Trump’s AI advisor, AI skepticism isn’t organic—it’s bankrolled by a $1 billion shadow network

• Baidu drops an open-source multimodal AI that claims to outperform GPT-5 and Gemini on vision tasks

📄 Visual Spatial Tuning 

📄 Too Good to be Bad: On the Failure of LLMs to Role-Play Villains ​​

📄 DeepEyesV2: Toward Agentic Multimodal Model  

📄 HaluMem: Evaluating Hallucinations in Memory Systems of Agents 

📄 Robot Learning from a Physical World Model

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on 𝕏!