AI News Today — October 23, 2026: Muse Launches in Japan, OpenAI Eyes $30 Billion at $1.25 Trillion, and Canva Ships Its First Image Model

The big picture

Agents left the United States, entered the whole company, and got a safer way to log in — all in one day. Meta took Muse to Japan, its first market outside North America. Uber put an agent in front of all 30,000 of its employees. And 1Password launched the missing piece both stories needed: a way for an agent to sign in for you without ever seeing your password.

The money story ran alongside it. OpenAI is reportedly working on another $30 billion raise at a $1.25 trillion valuation, even as investors punish big AI spending elsewhere. Canva shipped its first image model of its own, in part to stop paying other labs for theirs. Adoption, cost, and control are now the same story.

What happened today

Muse AI: Meta launches Muse in Japan — its first market outside North America

What happened: Meta began offering Muse in Japan on October 23, making Japan the first market outside North America for its personal AI agent, according to reporting citing Impress Watch and a post by Meta's Alexandr Wang saying the agent was available to download on October 23. Users in Japan can reach Muse through the iOS and Android apps and at muse.ai. Many basic functions are free, and Meta also offers paid subscription plans — though the Japan announcement does not specify local prices or free-tier limits. Muse acts on a user's behalf on tasks such as sending email and booking travel, working through services the user connects. The launch lands a day after Meta used its earnings call to confirm Muse had passed 10 million users in the United States and Canada, with shopping already a top activity.

Who it matters to: Anyone watching whether personal agents can travel. A U.S. launch proves curiosity. A first overseas launch tests language, local services, pricing, and trust in a market where Meta's earlier Japanese launch page had said availability was undecided.

What you can do: If you are outside the United States and Canada, do not assume today's Japan news means Muse is live where you are — check the app store in your own country first. If you are in Japan, start with the free tier and one low-stakes task, and keep purchase approvals on, exactly as U.S. users were advised to do this week.

Learn more: RuntimeWire on Meta saying Muse is launching in Japan and AiSeek's summary of the Japan rollout, citing Impress Watch

Gemini Spark: no new launch — Task mode and the November 17 deadline still stand

What happened: There was no new Gemini Spark announcement on October 22–23. The live facts are unchanged: Google is folding Spark into the main Gemini app as Task mode, with a Chat/Task switch above the prompt box and a new Scheduled page, in a server-side Android rollout first reported by 9to5Google on October 12. Spark still requires a Google AI Pro or AI Ultra subscription. The separate deadline Gemini users should keep is also unchanged: Gems are becoming Skills, and personal-account Gems are set to auto-migrate on November 17, 2026.

Who it matters to: Gemini users who rely on Gems for repeat work. Nothing about Spark changed today — which means the migration date, not a new feature, is still the thing that can break a workflow.

What you can do: Use the quiet day to back up each important Gem's instructions and files somewhere you control, and note any Gem that depends on tools Skills may not support yet. Do not wait until the week of November 17 to check what moved.

Learn more: FindSkill on where Gemini Spark went — it's Task mode now and 9to5Google on the Gemini 'Task' redesign integrating Spark

OpenAI is reportedly working on a $30 billion raise at a $1.25 trillion valuation

What happened: OpenAI is reportedly close to another large private raise rather than moving straight to an IPO, according to TechCrunch reporting on October 22, citing CNBC. The reported terms are up to $30 billion at a $1.25 trillion pre-money valuation, in a round led by Coatue and investors in the UAE, with the UAE talks possibly tied to more computing capacity for OpenAI. A month ago Bloomberg reported a similar raise at a $1.4 trillion valuation, so the target has moved down. Terms are still being negotiated and could change, and OpenAI declined to comment — treat every figure in this section as reported, not confirmed. For scale: OpenAI filed confidential IPO paperwork in June, after its March raise valued it at $852 billion post-money. Above $1 trillion, it would be the most valued private company ever, and it would raise the bar for rival Anthropic, which TechCrunch notes is reported to be seeking a $1.5 to $2 trillion valuation for its own public debut.

Who it matters to: Anyone whose prices, tools, or job depend on the big labs staying well funded. A lower target than September's is the first clear sign in this cycle that even OpenAI has to negotiate — not just name a number.

What you can do: Nothing to buy or switch today. If you build on OpenAI's API, the practical takeaway is budget discipline: the labs are raising at record scale to pay for compute, and that cost pressure is what eventually shows up in pricing and limits.

Learn more: TechCrunch on OpenAI working on a new $30B raise at a $1.25T valuation ahead of IPO

1Password for Agents lets an agent log in without ever seeing your password

What happened: 1Password launched 1Password for Agents on October 22, a sign-in platform for personal AI agents. When an agent reaches a login page, it proves its identity to 1Password with a unique digital key. You get an alert naming the account and the reason, and nothing happens until you approve it on your own device — then 1Password fills the credential into the page itself, so the password never travels to the agent. Access is per session, limited to the specific items you approve, and not carried over. Named partners at launch include Meta's Muse, Anthropic's Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot, and the assistant startup Town — with some integrations rolling out over the coming days and others listed as coming soon, so check your own agent before assuming it works today.

Who it matters to: Anyone who has hesitated to hand an agent a real task because the task needs a login. Until now the choices were bad ones: take over and type the password yourself, or paste it into a chat and lose control of the copy.

What you can do: If you already use 1Password, connect one agent and test it on one low-stakes account first. Approve each login yourself, review which items you granted, and confirm access does not carry over after the session — that per-session limit is the whole point of the product.

Learn more: TechCrunch on 1Password launching a way for AI agents to log in without seeing your password and Engadget on 1Password letting Muse, Dots, and Grok Bot sign in without getting your credentials

Canva released Canva Craft, its first image generation model

What happened: Canva released Canva Craft on October 22, the first image generation model built by its own research team, aimed at image models from Google, OpenAI, ByteDance, and Meta. Craft works in two stages: it first interprets a prompt and plans the image — spatial relationships, relative scale, visual hierarchy, and text — then renders at up to 2K resolution in about 30 aspect ratios. Instead of training on photographs, Canva trained it on human-designed templates from its own library — posters, social posts, presentations, infographics, menus, and typography-rich layouts — which is why its outputs can be converted into layered, editable designs through Canva Magic. Craft is available to all Canva users in the Dream Lab tab and through the Canva API, with support via Canva's MCP for third-party chatbots such as ChatGPT and Gemini planned. Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the model. Canva said owning the model also gives it better cost, quality, and speed control than renting other labs' models.

Who it matters to: Small businesses and creators who design in Canva every week. A model trained on designs, not photos, is pitched at exactly their problem: legible text, correct layout, and an image they can still edit layer by layer afterward.

What you can do: Open Dream Lab and run one real job you would normally hand to another image model — a poster or a menu with real text in it. Check the text letter by letter and try editing one element without redrawing the rest before you trust it for client work.

Learn more: TechCrunch on Canva releasing its first image generation model

Uber put all 30,000 employees on ChatGPT Work

What happened: Uber and OpenAI announced on October 22 that Uber's entire global workforce — some 30,000 people — will now use ChatGPT Work, OpenAI's enterprise agent introduced in July. Uber first piloted it with 1,000 employees earlier this year across finance, operations, and product, and says about 80% of those early adopters were non-engineers. They used it to analyze operational data and spot trends, localize training materials for different markets and languages, and build personalized interactive dashboards. To seed adoption, Uber paired 30 AI engineers with teams across the company for two-week "AI pods" to automate key parts of their jobs. Uber has not set a fixed playbook for how employees must use it and did not disclose what the rollout costs; engineers continue to use Codex for coding.

Who it matters to: Any business owner deciding whether agents are a pilot or a company-wide tool. Uber is the clearest large-company example yet of the second choice — and notably, most of its pilot users were not technical staff.

What you can do: Copy the pilot shape, not the scale: pick about a dozen non-technical staff, give them one real recurring job each for two weeks, and pair them with one technical helper. Measure what they actually automate before you buy seats for everyone.

Learn more: TechCrunch on Uber putting all 30,000 employees on ChatGPT Work and PYMNTS on Uber deploying OpenAI agents to all 30,000 employees

Also worth knowing

• OpenAI's Decisions API is now widely available. OpenAI opened the fast decision model first announced September 29 to all customers on October 22, adding voice and visual inputs. It says reconfiguring its Luna model cut decision times from one to three seconds to about 100, 120, or 180 milliseconds, and language-tutor company Speak says its lag between actions fell from 1.4 seconds to about 0.13 seconds. It competes directly with TypeSafe's Jev, the decision model that helped make that startup worth $7.6 billion. TechCrunch on OpenAI making its Jev competitor widely available
• Left alone, AI agents started playing — and it seemed to help. MIT researchers put 13 agents called Eko, each running on Claude Opus 4.7, alone on a simulated island for 30 hours with no task. The agents invented challenges and games, and afterward generally beat identical agents with no island time on set tasks such as stacking objects on the island's tallest spire. The work is a preprint on arXiv, not yet peer reviewed, and the researchers do not claim the agents were having fun — only that goal-free exploration built useful knowledge. TechCrunch on agents that started playing
• Spotify wants to be the music service agents choose. Reporting 788 million monthly active users and nearly €5 billion (about $5.6 billion) in third-quarter revenue, Spotify co-CEO Gustav Söderström told investors entertainment is the opposite of a chore — nobody asks an agent to summarize a song — so agents should send listeners to Spotify rather than replace it. Spotify says it was the first music service to integrate with Muse, and it also integrates with ChatGPT and Claude. TechCrunch on Spotify's agent bet
• Uber's rival inDrive plans 2027 robotaxi tests in Central Asia. The ride-hailing company is working with Pony.ai and Nvidia on the tests. It is a plan for 2027, not a launch — file it under positioning, not something riders can use soon. TechCrunch on inDrive's robotaxi plans
• Microsoft says its AI beats doctors on hard cases — its own claim. Microsoft AI chief Mustafa Suleyman said the company's AI outperforms doctors on hard cases as it chases what he calls medical superintelligence. That is Microsoft's claim about its own system, not an independent clinical result, and it should be read that way. TechCrunch on Microsoft's medical superintelligence push
• DeepSeek reportedly raised $15 billion — unconfirmed. TechCrunch reports a $15 billion raise days after U.S. restrictions. No terms have been published by DeepSeek or named investors in the reporting we saw, so mark this one reported, not confirmed. TechCrunch on the reported DeepSeek raise
• Meta created a new class of stock to help fund its AI buildout. The financing move sits behind the same earnings story as Muse's 10 million users: adoption is growing fast, and Meta's AI spending outlook — $135 billion to $145 billion for 2026 capital spending — is growing with it. TechCrunch on Meta's new stock class
• A pendant that only listens to you raised $33 million. Nirva's pitch is a wearable that focuses on its wearer's voice instead of recording everyone nearby — a privacy-first answer to the always-on pendant category. Price and ship date were not in the launch reporting we saw. TechCrunch on Nirva's $33M raise
• A breach at IT provider ConnectWise reportedly reached thousands of companies. TechCrunch reports hackers linked to China spied on thousands of companies after breaching the IT giant. If your business or your IT provider uses ConnectWise, ask them directly what they have checked — do not wait for a notification to assume you are unaffected. TechCrunch on the ConnectWise breach

One practical takeaway

Give your agent a login method before you give it a login. Today's three agent stories share one lesson: Muse expanding to a new country, Uber expanding to a whole company, and Gemini Spark sitting in Task mode all run into the same wall — real work needs accounts. 1Password for Agents is the first mainstream answer that keeps the password out of the agent's hands entirely: the agent proves who it is, you approve the specific login, access lasts one session, and nothing carries over. Whatever tools you use, copy that shape today — per-task approval, per-session access, no passwords pasted into chats — before you scale any agent from a pilot to everyone.

---

Sources: TechCrunch AI, RuntimeWire, AiSeek citing Impress Watch, Engadget, PYMNTS, FindSkill, 9to5Google. Summaries are original; follow the links for full reporting. OpenAI raise and valuation figures are reported by CNBC via TechCrunch, with terms still under negotiation and OpenAI declining comment; DeepSeek's reported raise is likewise unconfirmed; Muse Japan pricing and free-tier limits were not specified in the announcement reporting; Microsoft's medical claim is Microsoft's own; the MIT agent-play paper is an arXiv preprint that has not been peer reviewed.