AI research is turning into real-world results, not just demos. Anthropic put Claude to work designing more than 300,000 proteins to fight scorpion venom, while OpenAI's top models arrived on Google Cloud. Meanwhile the money and the warnings are getting louder, from a $6 billion valuation for an AI lab with nothing on sale to a leading researcher's claim that we have entered a singularity.

Summary: A research team spent three months prompting Claude to invent proteins that could neutralise scorpion venom, which harms over a million people a year. The model picked targets, set scoring thresholds and used a dozen design tools to generate more than 300,000 candidates. Of 211 designs tested in a lab, 17 bound to toxins at strong affinity. The work cost around $200,000 and compressed roughly a year of research into days.

Why it matters: This shows AI can now handle a big chunk of early drug discovery, opening the field to small teams rather than just large pharma labs.

Summary: On 21 October Google Cloud added three OpenAI models to its Model Garden: GPT-6 Astra for heavy coding and research work, GPT-6.1 Sol for close to the same quality at a lower cost, and GPT-6 Luna for fast everyday jobs. They are in private preview for selected enterprise customers, with general availability promised in the coming weeks. Usage is billed through Google Cloud Marketplace and counts toward a company's existing Google Cloud spending commitments. Developers can already connect Codex to the models.

Why it matters: Amazon and Microsoft already sell OpenAI's models, so Google is the last of the three big clouds to get them. It is now the only one offering the top models from OpenAI, Anthropic and Google in one place, so a team already on Google can test all three without signing a new contract.

Source: Techzine

Summary: An OpenAI model discovered two decade-old vulnerabilities in PHP, the language behind roughly 70 percent of websites. The bugs let attackers hijack servers without needing a login. Several European governments took sites offline as a precaution, and one infrastructure firm reported early signs of exploitation. The flaws are part of a wider set of issues still under investigation.

Why it matters: It is a double-edged moment: AI can now surface hidden risks in ageing code, but the same power sits in attackers' hands.

Summary: At its GM Forward event in New York on 21 October, GM unveiled an in-house OnStar AI Assistant. Drivers can change settings like the climate by voice and ask questions about the car's features. Connected cameras can switch on by themselves when the car senses trouble, such as a sharp honk of the horn, and save the footage to the OnStar app. It arrives first on the 2027 Cadillac Lyriq through a software update in the coming months, with more GM models to follow.

Why it is interesting: For decades OnStar was the button you pressed to reach a call centre. GM's product chief, Sterling Anderson, now calls it "a connected intelligence platform."

Source: TechCrunch

Summary: On 21 October developer Simon Edwardsson, co-founder of V7, released WorldTUI, a 360-million-parameter model that runs in your browser and imitates a computer terminal. There is no real computer behind it. The model guesses what the screen should show after each keypress, including the vim editor, tmux, a Python prompt and a game of snake. It learned from 336,000 recordings of real programs, trained in about 75 hours on a single gaming graphics card.

Why it is interesting: Edwardsson suggests typing a typo like "pythoon" to test it. A real terminal would reject it, so it's the quickest way to catch the model making things up.

Source: Simon Edwardsson

Summary: Baroness Campbell of Surbiton, who has spinal muscular atrophy, delivered a speech in the House of Lords using an AI-generated clone of her own voice. She called it her second self and said it lets her speak her own words. It is reportedly the first time an AI voice has been used in either House of Parliament. Her speech addressed personal health budgets in a health debate.

Why it is interesting: It shows AI used not to imitate others but to restore and preserve someone's own identity in a serious public setting.

Source: ElevenLabs

Context: Use this when you want a tough, structured critique of a business idea before committing time or money. It casts the AI as a strategy consultant running an investment committee review.

Prompt: Act as a high-level strategy consultant presenting my idea to a sceptical investment committee. Here is the idea: [describe it]. Rigorously critique it. Identify the biggest risks, the weakest assumptions, who the real competitors are, and what would have to be true for this to succeed. Then give a clear recommendation: proceed, revise, or drop, with reasons.

Example use case: A founder tests a subscription app concept and gets a blunt list of fatal flaws and must-fix assumptions before building anything.

Summary: Flapping Airplanes, a San Francisco AI research lab founded in late 2025, raised $1 billion at a $6 billion valuation, Bloomberg reported on 21 October. Index Ventures led the round with Lux Capital, joined by Sequoia, GV and Kleiner Perkins. The valuation is four times what it was in January, when the lab raised a $180 million seed round. It is working on ways to train AI with less data and computing power, and has no product on sale yet.

Why it is important: The lab hasn't published results showing its methods work, so investors are paying for a promise. If it delivers, cheaper training would lower costs for every business paying for AI.

Source: The Next Web

Summary: Speaking to financial leaders, an Anthropic co-founder argued AI progress has reached a singularity, citing exam scores jumping from 8 percent to nearly 68 percent on a hard benchmark in a year. He projected US GDP could reach $44 trillion by 2030 if adoption accelerates, and warned there is only a short window to decide whether to limit AI's intelligence.

Why it is important: The framing moves the debate from capability to control, pressuring governments to act while labs push ahead.

Summary: Several state attorneys general have sued OpenAI, alleging it misled parents about ChatGPT's safety for minors and that the product harmed children. OpenAI notes no system can be flawless and points to a teen version it launched voluntarily. Critics argue chatbots face no pre-launch safety checks despite heavy teen use.

Why it is important: It raises the question of whether AI products aimed at or used by children should face regulation similar to the toy industry.

Source: Reuters

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas