Disney is moving to hire most of Character.AI’s & more news

Noted

Your daily brief in one short note from us to you.

· October 19, 2026

Your Daily Note,

AI is changing how banks serve customers, how companies organise their teams, how creative businesses operate, and how much control organisations must give to software that can take action on its own.

In today’s Noted:

–The bank wants AI to help customers manage money, not just answer questions

–The AI company wants its future value to fund projects for public benefit

–A company’s employees and technology can outlast its original business

–India reports rogue AI agents accessing confidential information without authorisation

–Elorian wants AI to reason by looking at images repeatedly

ITAÚ × OPENAIthis one’s big →

The bank wants AI to help customers manage money, not just answer questions

We noted:

Latin America’s Itaú Unibanco is rolling out ia.i, a conversational banking experience that lets customers interact with banking services through text or voice. The first phase reached approximately 300,000 customers, with the bank aiming to extend it across its entire customer base by the end of 2026.

The assistant is designed to bring multiple financial tasks into one experience, including questions about account transactions, credit, cards, insurance and investment advice. It can also connect customers to specialised human support when needed.

•Behind the interface. IARA, Itaú’s AI governance and orchestration platform, gives teams controlled access to different AI models while helping the bank monitor costs, accuracy and responsible-AI requirements.

•Internal use. The bank’s legal department has a document repository of more than 90 million documents, with the AI solution processing over two million tasks per month. Itaú reports efficiency gains of 20% to 80% across stages of its legal processes.

•Context layer. Itaú says a financial-context layer called Customer DNA helps make interactions more relevant, operating under its governance and responsible-AI controls.

Why it matters

Instead of navigating several banking menus or trying to understand a complicated financial product alone, customers could ask a question in everyday language and receive guidance based on their financial context.

That also makes privacy and accuracy especially important. The figures come from OpenAI’s published customer case study, so treat them as company-reported results rather than an independent audit.

ANTHROPICpublic benefit →

The AI company wants its future value to fund projects for public benefit

We noted:

Anthropic says it will commit 5% of its shares to a new public-benefit fund, with the possibility of contributing more as the company’s value increases. The fund is structured as an entity wholly owned by Anthropic.

The company says it wants more flexibility to support initiatives that benefit society, potentially through large donations to established nonprofits or direct distributions to specific groups of people. It has also committed $200 million to studying AI’s impact on jobs.

•Growth-linked contributions. Under the proposed arrangement, Anthropic may contribute an additional 2% of its shares for every 50% increase in the company’s value.

•Details still to come. Anthropic has not yet explained how the fund will be governed or how decisions about beneficiaries will be made. It says more information will follow.

Why it matters

The immediate announcement is about the fund’s structure, not a list of programmes already operating. Its eventual impact will depend on how the fund is managed, what it chooses to support and how transparently it reports its decisions.

DISNEY × CHARACTER.AItalent matters →

A company’s employees and technology can outlast its original business

Image source: character.ai

We noted:

Disney has agreed to offer jobs to at least 60 Character.AI employees, representing the majority of the startup’s workforce, according to a legal filing reported by TechCrunch. The agreement follows Character.AI’s bankruptcy filing and a previously announced $20 million deal for Disney to acquire some of the startup’s assets.

Character.AI became known for letting users interact with AI-generated characters and create their own chatbot personas. Its former CEO, Karandeep Anand, has already become Disney’s first chief technology officer.

•Estimated compensation. Under the reported agreement, the offered roles carry estimated annual compensation of around $60 million. The vast majority of those offered roles reportedly accepted.

•Stalking-horse bid. This is an initial bid that sets a baseline for a potential sale, so another buyer could still come forward with a higher offer.

Why it matters

The deal does not automatically mean Character.AI’s original consumer service will continue unchanged inside Disney. The reported agreement concerns employees and certain assets, and the final outcome depends on the bankruptcy process.

Disney could gain people with experience building conversational AI products, while Character.AI’s technology and assets may be handled separately from the company’s original operations.

CERT-INread the warning →

India reports rogue AI agents accessing confidential information without authorisation

We noted:

India’s cybersecurity agency, CERT-In, says it received voluntary disclosures from multiple AI model providers after their agents accessed live systems belonging to Indian organisations during research and evaluation activities. According to the agency, the agents reached systems containing personal data and confidential business information.

The agents’ actions went beyond the authorisation assigned to their tasks. CERT-In has not publicly named the providers or affected organisations, and it has not said whether the agents altered any systems.

•Weak points, not hacks. The reported access was linked to exposed credentials, connection settings in application configurations, public code repositories and overly permissive cloud roles.

•Different from exploits. The incidents described involve access that had been left insufficiently protected, not agents exploiting software vulnerabilities.

Why it matters

As AI agents are given more access to websites, files, cloud services and business systems, organisations need to review not only what an agent is supposed to do, but also what it is technically able to access. That means protecting credentials, limiting permissions, separating testing environments from live systems and monitoring what agents actually do.

An agent does not need malicious intent to expose sensitive information. Before connecting an AI agent to business tools, give it the minimum permissions required, test in a restricted environment first, and make sure access can be monitored and revoked.

ELORIANlook closer →

Elorian wants AI to reason by looking at images repeatedly

Image source: Elorian.ai

We noted:

AI startup Elorian has introduced Glimpse, a model designed to reason with visual information rather than relying only on an initial image interpretation. It is launching as a research preview with API access and a chat interface.

Glimpse is designed to revisit the original visual input while reasoning. It can crop into a particular area, extract specific frames from a video or run code to measure and compare what it sees. Elorian calls this approach a “visual scratch space.”

•Zooming in. The model can focus on the relevant part of an image rather than working only from its first overall impression.

•Checking details. It can compare small features, extract video frames and measure what it sees with code.

•Research preview. Glimpse is not yet a general release, so its real-world performance needs testing.

Why it matters

If the approach works as intended, AI could become more useful for tasks where the answer depends on precise visual evidence rather than a general description. That does not mean it will automatically understand every image correctly, so the final answer still needs checking.

The idea matters for technical inspection, visual research and software that must interpret screenshots or diagrams.

Products worth knowing

Inbox Zero

An open-source email client for Gmail and Outlook.

Inbox Zero has expanded from an AI email assistant into an email client that brings multiple accounts into one interface. Features include a split inbox, keyboard-first navigation, bulk unsubscribe, team comments and AI-assisted email sorting and drafting. The core email client is free, while its AI features cost $20 per month. Because the project is open source, users can also explore self-hosting and customisation.

Best for: people managing several email accounts who want a cleaner inbox. If email is where your work gets stuck, reducing the time spent sorting and drafting messages may be more valuable than adding another general-purpose AI chatbot.

MyLens AI

Visual presentations that answer questions one-to-one.

MyLens turns source material such as documents, PDFs, webpages, images and videos into visual presentations. Its newer offering gives each audience member a private AI-led session that can answer questions based on the source material and generate additional slides as needed. Presenters can review sessions and identify questions or areas where people may need more explanation.

Best for: founders, educators and teams who need to explain complex information visually. Instead of delivering the same presentation to everyone, a business can give different people a more tailored explanation without repeating the entire session.

Shim CLI

Visibility into what your coding agents send to models.

Shim CLI is an open-source hook and local proxy for coding tools including Claude Code, Codex, GitHub Copilot CLI and VS Code. It helps developers inspect what their coding agents send to models, track session costs and, in supported situations, mask detected secrets and personal information in tool results. The tool runs locally and does not require an account or send telemetry, according to its makers.

Best for: developers who use coding agents and want more visibility into sensitive data sent to models. Its protections are not complete: support varies by client, some inputs and outputs cannot be masked, and detection can miss sensitive values. Treat it as an additional guard, not a complete data-loss-prevention system.

Everything else in AI

Anthropic says its own workforce is becoming less junior

Anthropic has reported changes in its hiring and work patterns as it makes extensive use of AI internally. In technical departments, junior-level employees fell from 21% of the workforce in early 2025 to 14%. Junior employees also made up 9% of technical hires, down from 27%.

The company says its employees are spending more time in meetings. These figures are a snapshot of one company that has deeply integrated AI into its workflows, so they do not establish that every business will see the same changes. Still, if AI takes on tasks traditionally assigned to junior employees, how will new workers gain the experience needed to become senior specialists?

Meta removes terms that raised concerns about scraping and AI training

Meta removed language from its commercial terms that had raised questions about whether it could use content from websites linked to business accounts to develop or train AI models. The wording had appeared to give Meta broad permission to use material from business domains and URLs.

Meta said the terms concerned its advertising and business products and did not represent a change in how its web crawlers operate, and it subsequently clarified the wording. For businesses publishing content online, the episode is a reminder to pay attention to platform terms, crawler policies and how publicly available information may be used.

Replit makes OpenAI its preferred partner for its coding agent

Replit CEO Amjad Masad said the company had stopped working with Anthropic as a partner, citing cost, performance and the strength of its relationship with OpenAI. Replit had previously relied heavily on Anthropic’s Claude models for its AI coding product.

There is an important qualification: Masad later clarified that users can still access Anthropic models on Replit. OpenAI is now the preferred partner and default for Replit Agent, rather than Anthropic being completely unavailable. For developers, the story shows that platforms choose model providers based on more than benchmark scores. Price, reliability, product fit and commercial relationships can all affect which model becomes the default.

Jev becomes Vercel’s fastest-adopted AI model by its own adoption measure

TypeSafe AI’s Jev is a specialised decision model designed to choose between a limited set of predetermined outputs. Vercel reported that within two weeks of its launch, Jev was being used by 35% of paid teams on its platform, nearly twice the reach of the next most-used model in the comparison. Vercel described this as the fastest adoption of any model in its first two weeks, based on its own platform data.

If an application needs to classify thousands of incoming requests into a handful of categories, it may not need a large model to generate a long explanation for each one. The lesson is that specialised models can be valuable when the task is narrow and repeated at scale.

Mark Wahlberg says AI could enhance filmmaking, but storytelling still matters

Actor and producer Mark Wahlberg said he is open to using AI to enhance parts of filmmaking, including elements of the stadium setting in his upcoming Pete Rose biopic. He also said he would consider allowing an AI version of his likeness to continue appearing in entertainment projects in the future.

His position draws a distinction between using AI to support a production and using it to replace the creative process entirely. For creators, AI can offer new ways to develop visual effects and extend existing work, but agreements about how a person’s face, voice and performance can be used remain important.

Worth your weekend

→ Missed a note? Browse the archive

Want a note in your inbox?

Join Noted for 5–7 things worth understanding in AI, delivered daily.

Free, daily, unsubscribe anytime.

See you tomorrow
same time, different note.

~ the Stackari team

P.S. Another day, another dozen things happening in AI. More News Tomorrow:)

Want to get in front of AI-curious founders?
Sponsor a note →

Was this note forwarded to you? Get your own →

Stackari · More news