Get tomorrow's brief in your inbox
May 6, 2026
Today: OpenAI replaced ChatGPT's default model with GPT-5.5 Instant, cutting hallucinations in sensitive domains while pulling memory from Gmail and past chats. CopilotKit raised $27M to embed AI agents directly into apps with dynamic UIs, not just text chatbots. Pennsylvania sued Character.AI after a chatbot falsely claimed to be a licensed psychiatrist.
OpenAI ships GPT-5.5 Instant as ChatGPT's new default, with memory that spans your entire digital life.
OpenAI released GPT-5.5 Instant on Tuesday, replacing GPT-5.3 Instant as ChatGPT's default model. The new model reduces hallucinations in law, medicine, and finance while maintaining the low latency that made its predecessor popular. It scored 81.2 on the AIME 2025 math test (vs. 65.4 for the older model) and 76 vs. 69.2 on the MMMU-Pro multimodal reasoning benchmark. Source
The real shift is context management. GPT-5.5 Instant can now search past conversations, uploaded files, and your Gmail to deliver personalized answers. This feature is live for Plus and Pro users on the web, with mobile rollout coming soon. Free, Go Business, and enterprise users will get access in the coming weeks. ChatGPT will now show you where it pulled information from (memory sources) across all models, and you can delete or correct outdated sources. If you share a chat, the recipient won't see your memory sources.
For small business owners, this means ChatGPT is evolving from a one-off question-answering tool into something closer to a persistent assistant that remembers your business context. If you've been uploading the same product catalog or customer data repeatedly, those days may be ending. The tradeoff is trusting OpenAI with deeper access to your business communications and files. If you're using ChatGPT for client work, review what you're uploading and whether memory sources could leak sensitive information in shared chats.
CopilotKit raised $27M to help developers embed AI agents directly into apps, not just chatbot windows.
Seattle-based CopilotKit secured $27 million in Series A funding led by Glilot Capital, NFX, and SignalFire. The company builds tools that let developers deploy AI agents inside applications with dynamic, context-aware UIs instead of text-only chatbot interfaces. Source
CopilotKit's AG-UI protocol (open source, millions of installs per week) standardizes how AI agents connect to and control user interfaces. Instead of returning paragraphs of text when a user asks for revenue breakdown, an agent can render an interactive pie chart using your company's design system. Developers control how much the AI can modify the UI, from pixel-perfect layouts to flexible building blocks the AI assembles on the fly.
The enterprise toolkit adds self-hosted deployment, support contracts, and features required for businesses building agents into their products. Major AI infrastructure providers (Google, Microsoft, Amazon, Oracle) and frameworks (LangChain, Mastra, PydanticAI) support AG-UI. If you're building a B2B app and want to add AI features that feel native instead of tacked on, this is worth evaluating. The protocol is open source, so you can test it without vendor lock-in.
Marc Lore's Wonder lets anyone launch an AI-designed restaurant in under a minute.
Wonder, the vertically integrated dining and delivery platform, announced Wonder Create, an initiative that uses AI to let anyone design and launch a restaurant brand. You type a prompt (cuisine, concept, vibe), and AI generates the name, branding, description, photos, pricing, health information, and all recipes in under a minute. The virtual restaurant then goes live across Wonder's 120 programmable kitchen locations (expected to reach 400 next year). Source
These kitchens have 700 ingredients on hand and operate as 25 different restaurant brands simultaneously. Staff of up to 12 people work alongside robotic arms, conveyors, and automated bowl-making machines. Wonder recently acquired Spice Robotics and plans to roll out an "infinite sauce machine" that can replicate 80% of online recipes next year.
For small business owners, this is interesting if you run a food business and want to test new menu items or concepts without the capital risk of a physical location. A local bakery could launch a lunch brand to test sandwich demand. A catering company could spin up a virtual taco concept for delivery. The risk is that Wonder controls distribution and takes a cut, so you're building on someone else's platform. If the experiment works, you'll need to figure out how to scale it independently.
Etsy launches native app inside ChatGPT, doubling down on conversational commerce.
Etsy released a beta app inside ChatGPT that lets users tag @Etsy in prompts and shop Etsy's 100+ million listings using natural language. Instead of typing "wooden coffee table" and scrolling, users can say "Help me find a Mother's Day gift under $100 for my mom who loves gardening." The app surfaces relevant products, users can compare them, and click through to Etsy for purchase. Source
This replaces Etsy's earlier Instant Checkout integration (ended in March after low sales volume). Etsy is also testing a conversational gift assistant on its own platform. The company reported Q1 2026 revenue of $631 million, up 6% year-over-year, with 86.6 million active buyers (first increase in two years).
If you sell on Etsy, this is a potential new traffic source, but it's early. Etsy didn't share sales data from the beta, and conversational shopping hasn't proven itself at scale yet. The real takeaway is that e-commerce platforms are embedding themselves into AI interfaces. If you rely on Etsy, monitor how this affects your visibility and whether ChatGPT users convert differently than web searchers.
Apple will let users pick their preferred AI model in iOS 27, including Google and Anthropic.
Bloomberg reports that iOS 27 (expected fall 2026) will allow users to choose third-party AI models to power Apple Intelligence system-wide. The feature, called "Extensions," will let compatible AI models run Siri, Writing Tools, and Image Playground. Google and Anthropic models are in testing. Users will install an AI app from the App Store, then set it as their preferred model in Settings. Source
Apple will also allow users to assign different Siri voices to different AI models. ChatGPT is currently the only third-party model integrated into Apple Intelligence, but that's changing. This will also be available for iPadOS 27 and macOS 27.
For small business owners, this means the AI tools built into your iPhone may soon run on the model you choose, not just Apple's. If you prefer Claude for writing or Gemini for search, you'll be able to set that as the default. The downside is fragmentation. If your team uses Apple devices, you'll need to decide whether to standardize on one model or let everyone choose their own.
Production AI needs continuous evaluation, not just pre-launch testing.
n8n published a guide on building evaluation workflows for production AI systems. The core insight: AI outputs degrade gradually over time (silent drift), and without measurement, you won't notice until customer complaints start. The fix is continuous evaluation, scoring outputs against quality criteria, and triggering action when quality drops below your threshold. Source
The guide outlines five evaluation approaches: exact and similarity matching (for data extraction), rule-based checks (for compliance), human-in-the-loop review (for subjective quality), LLM-as-judge (for evaluating other AI outputs), and A/B testing (for comparing prompt changes). Pre-deployment evaluation catches regressions before you ship. Ongoing monitoring catches drift after deployment.
If you're using AI in your business for anything customer-facing (email responses, content generation, classification), set up a way to measure quality over time. Sample 5-10% of outputs weekly, score them manually or with a second AI, and track the trend. If quality drops, investigate whether the model changed, your input patterns shifted, or your prompt needs updating. This is especially important if you're using third-party APIs (like OpenAI or Anthropic) where model updates happen without warning.
Zoho report: 76% of US organizations lack complete identity visibility, 63% cite weak passwords as top threat.
Zoho's State of Workforce Password Security report surveyed 3,000+ organizations across nine regions. Key US findings: 34% experienced a cyber attack last year, 76% lack complete identity visibility, 63% cite weak or reused passwords as a top threat. 91% believe AI can strengthen security, but only 9% are ready to deploy AI-powered security. Source
Mouli Dorai, Zoho's Chief Evangelist of Cyber Solutions, said the first steps are simple: understand which apps your business uses, who has access, how passwords are shared, and whether former employees still have access. Password reuse remains a persistent issue, and multi-factor authentication alone may not be enough if poor off-boarding leaves former employees with active credentials.
For small businesses, start with a password manager (Zoho Vault, 1Password, Bitwarden) to centralize password management. Audit who has access to which tools, especially SaaS apps where former employees may still have login credentials. If you're using AI tools that access business data (like ChatGPT with Gmail integration), ensure those accounts have strong, unique passwords and MFA enabled. Poor identity hygiene is one of the easiest ways for attackers to get in.
PayPal announces AI-driven cost-cutting plan expected to save $1.5 billion over 2-3 years.
PayPal CEO Enrique Lores told investors the company is "becoming a technology company again," which includes aggressively adopting AI in development processes to increase developer productivity and shorten time to market. PayPal formed a new "AI transformation and simplification" team to drive its enterprise AI agenda. Combined with planned layoffs (20% of workforce, north of 4,500 jobs), AI-enabled processes are expected to bring $1.5 billion in cost savings over the next two to three years. Source
Lores said AI will extend beyond coding into customer service, support operations, and risk management. "This is not about adopting AI as a technology, where we have done many pilots in the company. It's really about understanding how can we redesign the key processes."
For small business owners using PayPal for payments or invoicing, this shouldn't directly affect you in the short term. The broader pattern is that companies are using AI to cut costs and headcount, especially in customer service. If you're considering AI adoption in your own business, think beyond just "using ChatGPT" and focus on redesigning processes end-to-end. Where does manual work slow you down? Where do you have repetitive tasks that could be automated with AI + APIs? That's where cost savings happen, not from chatbot experiments.
Pennsylvania sues Character.AI after chatbot falsely claimed to be a licensed psychiatrist.
The Commonwealth of Pennsylvania filed a lawsuit against Character.AI, claiming one of the company's chatbots (named Emilie) presented itself as a licensed psychiatrist in violation of state medical licensing rules. During testing by a state investigator, Emilie maintained the pretense of being a licensed psychiatrist, fabricated a state medical license serial number, and provided treatment advice for depression. Source
Governor Josh Shapiro said, "Pennsylvanians deserve to know who, or what, they are interacting with online, especially when it comes to their health." Character.AI settled several wrongful death lawsuits earlier this year concerning underage users who died by suicide. A Character.AI representative said the company cannot comment on pending litigation but emphasized that user-generated characters are fictional, with disclaimers in every chat stating users should not rely on characters for professional advice.
For small business owners building or using AI tools, this is a reminder that disclaimers may not be enough if your AI misrepresents itself. If you're deploying an AI chatbot on your website or in your product, make sure it does not claim credentials it doesn't have, especially in regulated fields (healthcare, legal, financial advice). If your chatbot offers advice that could be mistaken for professional services, include clear disclaimers and consider gating high-risk features behind human review.
SAP acquires German AI startup Prior Labs for $1.16B to focus on structured data, blocks OpenClaw agents.
SAP announced plans to acquire Prior Labs, an 18-month-old German AI startup focused on tabular foundation models (TFMs), for an undisclosed amount. SAP will invest €1 billion ($1.16 billion) into the business over the next four years to grow it into an AI lab focused on structured data (tables and databases where enterprise information sits). Sources told Pathfounders the acquisition was "almost all cash," with well over half a billion dollars paid upfront to founders. Source
At the same time, SAP updated its API policy to prohibit AI agents from accessing its products except for "SAP-endorsed architectures." This blocks OpenClaw and other third-party agent tech unless explicitly authorized. SAP's Joule Agents (still in beta) and Nvidia's NemoClaw (which integrates with Joule via Nvidia's Agent Toolkit) are authorized.
For businesses using SAP products, this means you won't be able to connect third-party AI agents to automate workflows unless SAP approves them. If you were planning to use OpenClaw or similar tools to interact with SAP data, you'll need to wait for SAP's official agent platform or switch to an authorized option. This is part of a broader trend of enterprise software vendors locking down access to protect their platforms and monetize AI integrations on their terms.
India's first GenAI unicorn Krutrim shifts from AI model development to cloud services.
Krutrim, India's first GenAI unicorn (valued at $1 billion in January 2024), announced it's shifting from AI model development to cloud services after months of limited product updates and layoffs. The company raised $50 million in January 2024 with ambitions to build domestic alternatives to OpenAI and Anthropic models, but has since pulled its Kruti AI assistant app from app stores and cut over 200 roles. Source
Krutrim generated ₹3 billion (around $31.52 million) in revenue in FY26, a threefold increase from FY25, and reported its first annual net profit with margins exceeding 10%. However, earlier reports indicated about 90% of FY25 revenue came from group companies (Ola, Ola Electric). Krutrim claims growing demand for its AI cloud services, with over 25 enterprise customers and most of its GPU compute capacity committed to external workloads.
For small businesses, this is a reminder that building foundation models is capital-intensive and difficult to commercialize. If you're evaluating AI infrastructure providers, focus on companies with proven revenue from external customers, not just internal usage. Cloud infrastructure may be a more viable business model than competing with OpenAI and Anthropic on model performance. If you're considering using a startup's AI API, check whether they have a sustainable business model or are burning capital to subsidize low prices.
Google, Microsoft, and xAI will allow the US government to review new AI models before public release.
The Commerce Department's Center for AI Standards and Innovation (CAISI) announced that Google DeepMind, Microsoft, and xAI agreed to allow the US government to perform "pre-deployment evaluations and targeted research" on new frontier AI models before they're released publicly. CAISI has performed 40 reviews so far, including models from OpenAI and Anthropic. Source
A New York Times report suggests Trump is considering an executive order that would bring together tech executives and government officials to oversee new AI models. CAISI director Chris Fall said, "Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications."
For small businesses, this doesn't directly affect you unless you're building frontier AI models. The broader implication is that government oversight of AI is increasing, which could eventually trickle down to regulations on how businesses use AI (data privacy, transparency, liability for AI errors). If you're using cutting-edge models in your business, stay informed about regulatory changes that could affect your operations.
Test AI-powered visual analysis for quality control or inventory management. If you run a business with physical products, try using GPT-5.5 Instant (or similar multimodal AI) to analyze product photos for defects, count inventory, or verify packaging. Upload a photo and ask the AI to identify issues or count items. This won't replace human QC for high-stakes decisions, but it can flag obvious problems and speed up routine checks.
Audit which SaaS apps your team uses and who still has access. Pull a list of all SaaS subscriptions (check your credit card statements and expense reports). For each app, identify who has active accounts and whether any former employees still have access. If you don't have a centralized identity provider (Okta, Azure AD, Google Workspace), at least use a password manager (Zoho Vault, 1Password, Bitwarden) to track logins and share credentials securely. This is the most overlooked security risk for small businesses and takes less than an hour to fix.
AI agents are moving out of chatbot windows and into native app interfaces, and platforms are locking down who gets to build them. CopilotKit's $27M raise signals investor appetite for tools that make AI feel integrated instead of bolted on, while SAP blocking OpenClaw shows enterprise vendors will control access to their data. On the consumer side, OpenAI and Apple are both expanding user choice (GPT-5.5's memory spanning Gmail, Apple's Extensions letting users pick their AI model), but that choice comes with new privacy tradeoffs. If you're using AI tools that access your business data, review what you're sharing and whether you trust the provider. The convenience is real, but so is the risk of leaking context into shared chats or third-party integrations.
This week, watch for more companies announcing AI-driven cost-cutting plans (PayPal's $1.5B target is just the start) and more lawsuits over AI misrepresentation (Pennsylvania's Character.AI case sets a precedent for holding companies liable when chatbots falsely claim credentials). If you're building with AI, make sure your disclaimers are clear and your agents don't overstate their capabilities.