Welcome back {{firstname}}! Nvidia just released the first open-source AI models built specifically for quantum computers, OpenAI launched a cybersecurity model that takes a direct shot at Anthropic's restricted Mythos rollout and Chrome is turning your best prompts into reusable one-click shortcuts.
In today’s Generative AI Newsletter:
Quantum AI: Can Nvidia's open-source Ising models solve the problems keeping quantum computing stuck in the lab?
Cyber defence: Why is OpenAI opening its security model to thousands while Anthropic locks Mythos behind a 40-org whitelist?
Claude Code: Is Anthropic building a command centre for half-human, half-AI dev teams?
Chrome Skills: Could saved prompts in Chrome finally make browser-based AI feel useful?
Latest Developments
Nvidia Releases Open-Source AI Models Built for Quantum Computers
Nvidia has released Ising, a family of open-source AI models designed to work directly with quantum hardware. The models target two of the biggest technical barriers to scaling quantum computing: machine calibration and real-time error correction.
The details:
Calibration automation: The first Ising model automates quantum machine tuning, cutting a process that previously took days down to hours.
Error correction: The second model fixes errors during computation, running 2.5x faster and 3x more accurately than the best existing open-source alternative.
Institutional backing: More than 20 institutions are using Ising at launch, including Harvard, Cornell, Fermilab, IonQ and UC Santa Barbara.
Jensen's framing: Nvidia CEO Jensen Huang described Ising as "the operating system of quantum machines," positioning AI as the critical infrastructure layer the technology has been missing.
This is the same playbook Nvidia ran with self-driving cars and robotics: release the open AI layer, lock in the ecosystem, own the infrastructure underneath. The quantum computing market is projected to hit $11 billion and Nvidia is planting its flag before the serious competition arrives. If Ising delivers on calibration and error correction at scale, Nvidia won't just be supplying GPUs to quantum labs. It will be defining how quantum computers actually work.
Special highlight from our network
Build faster with Claude Code, without guessing what works.
In 6 live sessions, you’ll build your own claude.md system and learn a clear 13-step workflow to go from idea to shipped code. Use real examples, reusable skills and proven patterns across development, testing and docs.
No theory-heavy lectures. No prompt guessing.
Just a system you can use from day one.
OpenAI Launches GPT-5.4-Cyber to Counter Anthropic's Restricted Mythos Rollout
OpenAI has introduced GPT-5.4-Cyber, a version of its flagship model built for defensive cybersecurity work. The timing and framing are deliberate: this is a direct response to Anthropic's Mythos release, which restricts access to roughly 40 trusted partner organisations.
The details:
Open access model: GPT-5.4-Cyber is available to anyone who passes identity verification through OpenAI's Trusted Access for Cyber initiative. No whitelist, no partner programme.
Reverse engineering capability: The model can analyse compiled software to identify malware and security vulnerabilities without needing access to original source code.
Philosophy clash: OpenAI researcher Fouad Matin called cyber defence a "team sport" and argued against restricting who gets to defend their systems.
Government attention: Treasury Secretary Bessent convened an emergency briefing with Wall Street leaders last week over concerns about Mythos' hacking capabilities.
The two biggest AI companies are now taking opposite approaches to the same problem. Anthropic thinks restricting access to its most powerful cyber model protects the ecosystem. OpenAI thinks arming as many defenders as possible is the better bet. Neither approach is obviously right. Mythos posted monster benchmark scores and has already demonstrated real-world hacking capability that alarmed government officials. Giving thousands of organisations a less powerful but more accessible alternative could still shift the balance toward defenders, provided GPT-5.4-Cyber actually performs at the level needed.
Anthropic Redesigns Claude Code Around Multi-Session AI Development
Anthropic has overhauled its Claude Code desktop app with a redesign built around the reality that developers now run multiple AI sessions at once. Alongside the visual refresh, the company launched Routines, a feature that automates recurring development tasks.
The details:
Session management: A new sidebar displays all live and recent coding sessions, with filters by status or project and auto-archiving when a pull request is closed or merged.
Customisable workspace: Drag-and-drop panes let developers monitor multiple sessions, edit files, run tests and preview output without leaving the app.
Routines: A new research preview that runs AI tasks on a schedule, via API or when specific GitHub events occur. Configure once, run automatically.
Integrated tooling: The app now includes a built-in editor and terminal, removing the need to switch between Claude Code and external development tools.
Anthropic is betting that the developer's role is shifting from writing code to managing AI agents that write code. The multi-session layout and Routines both point in the same direction: a future where a single developer oversees several AI workers running in parallel. Whether that future arrives this year or in three years, Anthropic is building the interface for it now.
Chrome Turns Your Best Prompts Into Reusable One-Click Shortcuts
Google has launched Skills in Chrome, a feature that lets users save AI prompts and reuse them as one-click shortcuts inside Gemini in Chrome. The update also includes a curated library of pre-built Skills for common tasks.
The details:
How it works: Open Gemini in Chrome and type a forward slash to access your saved Skills or browse the library. Each Skill runs a saved prompt instantly.
Use cases: Scanning documents, calculating calories, comparing products across tabs or any repetitive prompt you find yourself typing regularly.
Tab integration: Users can attach open browser tabs to prompts using the @ symbol, letting Gemini read and compare content across multiple pages.
AI Pro upgrade: Google AI Pro subscribers get access to Chrome automations, where Gemini can browse the web and execute tasks on your behalf.
The gap between a chatbot and a genuine productivity tool has always been the friction of re-prompting. Skills removes that for Chrome users. The real test is whether people actually build and maintain their own prompt libraries or whether the pre-built ones carry all the weight. Google is clearly betting on both, but the curated library will likely drive more adoption than DIY shortcuts.
Tool of the Day: AnySpeech
AnySpeech converts text into natural-sounding speech and lets you create voiceovers in seconds. It supports multiple voices and languages, and the output quality is a clear step above most browser-based TTS tools.
Try it yourself: Head to AnySpeech, paste in any text (a blog intro, a script draft, a newsletter paragraph) and generate audio. Test a few different voice options to find one that suits your use case. If you produce any kind of content that could work as audio, this is worth five minutes of experimentation.
Light Bytes
New Anthropic model incoming: Rumours are circulating that Anthropic could release a new model as soon as this week. It is reportedly not Mythos.
Baidu's compact image generator: Baidu released ERNIE-Image, an 8 billion parameter open-weight text-to-image model that approaches top-tier rivals on benchmarks despite its relatively small size.
Hiro winds down, team joins OpenAI: AI personal finance startup Hiro announced it is shutting down operations, with its staff moving to OpenAI.
AWS enters drug design: Amazon launched Bio Discovery, an agentic platform for antibody drug design that includes biological foundation models and a built-in lab network for synthesis and testing.
ChatGPT hits 1 billion weekly users: OpenAI co-founder Greg Brockman claimed approximately 1 billion weekly users across ChatGPT and Codex, marking the company's 10th anniversary.
Enterprise vibecoding goes secure: Superblocks 2.0 brings enterprise-grade security controls to AI-generated app development, letting employees build while IT teams maintain oversight and audit trails.
