Overview
AI moved further into coding, creative work and everyday products today. Anthropic released cheaper, more capable models, Google introduced new tools for image and video work, and Dyson put computer vision inside a toothbrush. Elsewhere, Wall Street questioned legal billing in the AI era, while X continued its push into personal finance.
The big picture
Fable 5.1 takes on longer coding jobs at a lower cost
Anthropic released Fable 5.1 and Mythos 5.1, with the former now live in Claude Code and the Claude Platform. Fable 5.1 can run for longer before asking for help, reports when it is stuck more clearly and produces less rigid writing. API cache reads are also 75% cheaper, while agentic coding tasks are expected to cost roughly half as much as with Fable 5.
Early experiments included 100 distinct HTML files from a single prompt and a playable 3D racing game. Those demos need more scrutiny than a benchmark, but they show why developers are interested in the model’s ability to carry larger projects through without constant intervention.
Fable’s system prompt was reportedly extracted within an hour
The launch was quickly followed by a security headache. Jailbreak researcher Pliny reportedly published more than 270,000 characters from Fable 5.1’s system prompt, revealing internal instructions covering tools, memory, search, copyright and safety controls.
A system prompt leak does not expose model weights, but it can help researchers and attackers map guardrails or craft more focused jailbreaks. The speed of the extraction also underlines how little time model makers have before a new release is tested from every angle.
Dyson puts a camera and dental AI inside a toothbrush
Dyson’s £499 CameraJet combines brushing, mouth imaging and automatic flossing. A tiny camera captures 28 images per second, while an AI system trained on 470,000 dental images identifies gaps and fires targeted mouthrinse jets within 100 milliseconds.
Processing happens on the device, with live images available through a phone app. It is an ambitious piece of bathroom hardware, though its price and the prospect of watching your own teeth in close-up may divide opinion.
Google Pics brings object-level AI editing to Workspace
Google Pics is a standalone image creation and editing app for Workspace customers and Google AI Pro and Ultra subscribers. It can generate images, edit individual objects, rewrite or translate text inside an image, upscale work to 4K and pass assets into Docs and Slides.
The focus is control rather than repeatedly rewriting prompts and hoping for the right result. Team collaboration is built in, although Google’s growing collection of image products, including Imagen and Gemini, has already prompted some confusion about where Pics fits.
Gemini learns to inspect long videos more selectively
Google has added agentic video understanding to its latest Gemini models. Instead of sampling a long video at a fixed rate, the model can decide which frames, audio segments and transcript passages deserve closer inspection.
Google reports token use falling by as much as 88%, costs dropping by 66% and accuracy rising by 7% in its tests. The feature is available through the Gemini API and Google’s enterprise agent platform, with consumer and YouTube integrations planned.
Wall Street wants AI savings reflected in legal bills
Large banks are reportedly pressing law firms to cut fees for routine work now handled faster with AI, including document review and research. Fixed prices and competitive bids are gaining favour over charging for every associate hour.
Banks still appear willing to pay for senior judgement on complex deals and disputes. The pressure is instead aimed at a billing model that rewards time spent, even when software can complete part of the work in minutes.
X Money shows its plan for an account inside the social app
X design lead Benji Taylor shared a short look at X Money, showing balances, transaction history, revenue-share payments and separate main, savings and freelance accounts. The interface also includes controls for sending, requesting, depositing and transferring money.
The demo lists a 6% APY savings account and arrives during a phased US rollout to Premium subscribers. X has talked about becoming a home for both conversation and payments for years, and this gives that ambition a more concrete form.
A single Grok agent becomes the front desk for everything else
Farzad shared a Grok setup built around a master agent that handles every request, delegates work to specialist agents and keeps the rest of the system hidden. In the example, the main bot manages 14 other bots while presenting only a running action list and the decisions that need human input.
It is a practical answer to agent overload. Rather than juggling separate chats for work, administration and personal tasks, the operator talks to a single coordinator and lets it decide where each job belongs.
GitHub CLI can now attach images and video directly
GitHub CLI has added a repeatable --attach flag for local images and videos. Developers can upload media while creating an issue, pull request or comment, then reference it directly in the body.
It is a modest update that removes an annoying detour through the browser, particularly when bug reports need screenshots or short recordings. The feature is available across all GitHub plans.
Sam Altman discusses slowing frontier research and the AI backlash
Alex Heath’s two-part interview with Sam Altman covers OpenAI’s next model, Astra agents, alignment concerns, regulation and competition with Anthropic. It also addresses the harder public questions around creative jobs, wider displacement and whether the current spending boom can last.
The notable theme is restraint. OpenAI may be racing competitors, but Altman also discusses reasons to slow frontier research and give safety work more room, an argument that will be judged against the company’s next releases.
