AI Assembled 040626

Welcome to AI Assembled, the top AI news for small business.

I'm Chris Daley, and each week I bring you the most important AI developments

that can transform your business. Let's dive into this week's top stories.

On March 31st, a 135-page class-action lawsuit landed in federal court in San Francisco,

alleging that perplexity AI, the search and research chatbot valued at $20 billion,

had been secretly sharing users' private conversations with Meta and Google.

The lawsuit names all three companies as defendants.

The plaintiff alleges that tracking technology embedded in perplexity's code

transmitted the full text of user chats, the questions and the AI's responses

directly to Meta and Google, along with identifying information, including email addresses,

Facebook IDs, IP addresses, and device details.

The detail that has alarmed privacy lawyers most is this.

The tracking allegedly continued even when users activated perplexity's incognito mode,

a feature the company explicitly markets as creating anonymous threads that expire after 24

hours and are not saved. The lawsuit calls that claim a sham.

The implications for small businesses are immediate and concrete.

Think about what your team actually does in perplexity.

If employees have used it for legal research, financial planning, competitive analysis,

HR questions, or anything involving sensitive client or customer data,

that information may have been transmitted to Meta and Google's advertising infrastructure

without consent. Perplexity spokesperson told reporters they had not been served the lawsuit

and could not verify its claims. Here is what small businesses should do right now.

First, audit how your team uses perplexity and what kinds of questions they ask.

Second, consider temporarily restricting its use for sensitive work until the case develops.

Third, and this applies to every AI tool you use,

read the actual privacy policy in terms of service before putting confidential business

information into any chat interface. The comfortable conversational tone of these tools

creates a sense of privacy that the technology does not always support.

This case is a reminder that the interface is not the policy.

The story that sent Alphabet's market valuation above $4 trillion this year

is now rolling out to your iPhone. Apple has handed Siri to Google.

Under a multi-year agreement announced in January and now entering active deployment,

Apple's Siri will be powered by Google's Gemini 3 models, the same technology that drives

much of Samsung's Galaxy AI. The deal covers Siri and other Apple intelligence features across

Apple's installed base of more than two billion active devices. This is a significant reversal.

For the past year, Apple had integrated OpenAI's chat GPT into its devices,

allowing Siri to tap chat GPT for complex questions. Apple has now concluded that Google's

technology provides a more capable foundation and a more privacy-compatible one,

since Google's arrangement keeps Apple user data within Apple's own private cloud compute

infrastructure. Financial terms were not disclosed, but Bloomberg reported Apple is paying

approximately $1 billion a year for the right to use Google's AI.

For small businesses, the practical implications unfold on two levels.

In the short term, Siri is about to get dramatically more capable.

If your team uses Apple devices and relies on Siri for voice-driven productivity,

scheduling, dictation, research, workflow shortcuts, the quality of those interactions

is going to improve significantly over the course of this year. That is straightforwardly

good news for daily workflows. On a strategic level, this deal reshapes the AI

competitive landscape in ways that matter beyond any single product.

Google now powers the default AI assistant on both the world's largest Android ecosystem

and on Apple's devices. That is an extraordinary concentration of AI

distribution for a single company. For small businesses that have built marketing or discovery

strategies around AI search, which is increasingly how customers find products and services,

the consolidation of AI infrastructure into Google's Gemini platform

reinforces the importance of being visible in Google's AI results, not just in traditional

search rankings. If you want to know whether your business is optimized for AI search discovery,

there is a free grader that takes about five minutes and tells you exactly where you stand.

Anthropic had what can only be described as a double whammy week, and both events have direct

implications for any small business using their tools. The first happened on March 31st.

An NPM packaging error caused Anthropic to accidentally publish the entire source code of

Claude code, its flagship AI coding agent, to the public NPM registry. The file was a 59.8

megabyte JavaScript source map containing approximately 512,000 lines of TypeScript.

It was discovered by a developer at 4.23 in the morning, posted publicly on social media

and mirrored across GitHub within hours. Anthropic confirmed the incident and called it a

release packaging issue caused by human error, not a security breach. No customer data or

credentials were exposed, but the significance of the leak goes beyond the accident itself.

Claude code generates an estimated $2.5 billion in annualized recurring revenue for Anthropic.

What leaked is essentially the engineering blueprint for that product, including a sophisticated

three-layer memory architecture, an autonomous background processing system called Kairos,

and a feature described in the code as undercover mode, which allows Claude code to make contributions

to open source repositories without revealing that an AI was involved. Anthropics competitors now

have a detailed roadmap for how to build a product that has been two years in the making.

There is also a security action item. On the same day as the leak, a separate malicious

package was distributed through NPM targeting the Axios library, versions 1.14.1 and 0.30.4 containing

a remote access Trojan. If anyone on your team installed or updated Claude code via NPM on March 31st

between midnight and 3.30 in the morning UTC, check your package lock files immediately for

those Axios versions. If found, treat the machine as compromised, rotate all API keys, and perform

a clean reinstall. The second Anthropic story arrived on April 4th. Effective that day, Claude

subscriptions no longer cover usage of third-party tools, specifically OpenClaw. The popular AI

agent platform that tens of thousands of small businesses were using to manage inboxes,

calendars, and automated tasks now requires a separate pay-as-you-go API key.

Subscribers received a one-time credit equal to one-months plan cost.

The OpenClaw creator, who had been hired by OpenAI for millions of dollars, said he tried

to talk Anthropic out of the change and succeeded only in delaying it by one week.

If your team used OpenClaw on a Claude subscription, your costs have changed as of Saturday.

On Thursday, April 2nd, OpenAI announced it had acquired TBPN, the technology business programming

network, a live daily three-hour talk show that airs on YouTube and X and has built one of the

most loyal audiences in Silicon Valley. The show is co-hosted by former tech founders John

Cougan and Jordi Hayes, and it functions something like a sports center for the tech industry,

a place where major CEOs, including Mark Zuckerberg, Satya Nadella, Mark Benioff,

and Sam Altman himself, have come to discuss the news and occasionally make it.

This is OpenAI's first acquisition of a media company, and the financial terms were not disclosed.

TBPN was on track to generate more than $30 million in revenue in 2026.

The show will continue to operate under its own brand, with the host retaining editorial

independence, their word, and it will be housed within OpenAI's strategy organization,

reporting to Chris Lahain, the company's chief political operative.

Let's sit with that structure for a moment because it matters.

The hosts have said they will choose their guests and make their own editorial decisions.

Sam Altman has said he does not expect them to go any easier on OpenAI.

Those assurances may be genuine, they are also exactly what you would expect any party to say

at the time of an acquisition announcement. What this story actually tells small businesses

is something larger about the information environment you operate in.

When AI companies begin acquiring the media channels that cover them,

the boundary between journalism and advocacy becomes something every informed consumer of

tech news needs to keep in mind. OpenAI is not the first company to pursue this strategy,

and it will not be the last. For business owners specifically, TBPN has been a valuable source

of candid insider-level reporting on the AI industry. It has featured honest conversations about

AI failures, competitive dynamics, and product limitations. Whether that continues under OpenAI's

ownership is a legitimate open question. Pay attention to which guests appear and which stories

get the full treatment. The answer will tell you more than any press release about what editorial

independence actually looks like in practice. On April 1st, Vancouver-based Sanctuary AI

released a demonstration that the robotics industry will be studying for weeks.

The company's five-fingered hydraulic robotic hand repeatedly reoriented a lettered cube,

picking it up, repositioning it, and setting it back down 10 consecutive times without dropping it.

That sounds simple. In robotics, it represents a benchmark that has alluded researchers for years.

What makes the demonstration significant is not just the physical dexterity. It is the method behind it.

Sanctuary AI achieved what engineers call zero-shot SIM-to-real transfer.

The AI policy controlling the hand was trained entirely in a virtual simulation.

It had never been tested on physical hardware before the moment it was deployed in the demo.

That it worked, cleanly, repeatedly, under real-world physics, is a genuine milestone.

The gap between what robots can do in digital training environments and what they can do in physical

reality has historically been enormous. Sanctuary AI is demonstrating that gap is closing.

This is relevant to small businesses for reasons that are closer than most owners realize.

The robotics industry is not only competing on the high end,

Tesla's Optimus, Figure, and Boston Dynamics targeting automotive and logistics giants.

The mid-market is coming. Small manufacturers, warehouse operators, food processing companies,

and commercial cleaning services are all within the scope of where humanoid

and dexterous robotic systems will reach within the next three to five years.

Sanctuary AI's demo joins a week in which BMW confirmed ongoing deployment

of its eon humanoid robot at its Leipzig plant, where a prior US pilot had moved 90,000

components over 1200 operating hours. The pattern is consistent.

Physical AI is advancing faster than the mid-market is preparing for.

If you run any kind of operation that involves repetitive physical tasks and you have not yet

looked seriously at what automation might mean for your cost structure, that research is overdue.

The same AI infrastructure underpinning these robots is reshaping how your customers find you,

how they evaluate you, and how they decide whether to buy from you.

Find out how your digital presence looks to AI systems

by going to freshmediaworks.com slash AI ready.

Free five minutes immediate clarity on where you stand.

That URL again, freshmediaworks.com slash AI ready.

Let's close this week with something practical. Six stories, each with a specific ask.

The perplexity lawsuit asks you to audit what your team puts into AI chat tools.

The lawsuit's core allegation that private conversations were routed to meta and google

without consent, even in incognito mode, is a reminder that AI tools are not neutral containers.

They have business models. Those business models may involve your data.

Before your next team meeting establish a clear policy, which AI tools are approved for

sensitive work and what kinds of information should never go into a chat interface.

The Apple Google Gemini deal asks you to update your mental model of who owns the AI

assistant layer on mobile devices. The answer increasingly is Google.

If your business relies on being findable through search, through voice, through AI-powered

recommendations, the consolidation of AI infrastructure into Google's Gemini platform is not

background noise. It is infrastructure. Treat it like one. The Claude code leak and the open

Claude pricing change both ask the same thing. Know your tools and know their vulnerabilities.

If you use Claude code, check your package files for the malicious Axios versions flagged

this week. If you use OpenClaw, your billing structure changed on April 4th. Review your accounts.

In both cases, the lesson is that AI tools require active management, not passive consumption.

The OpenAI acquisition of TBPN asks you to think critically about where your information comes

from. When AI companies own the media that covers them, the responsibility for evaluating that

coverage shifts to you. Diversify your new sources. Read coverage from outlets that have no

commercial relationship with the companies they're reporting on. And the sanctuary AI demo asks you

to look at your physical operations with fresh eyes. The timeline for affordable, capable

robotics in mid-market settings is accelerating. That is not a reason to panic. It is a reason

to plan. One more action you can take today that costs nothing. Find out how your business looks

to AI search and AI assistance. Go to freshmediaworks.com slash A. I, ready for the free A. I,

optimization website, grader. It takes five minutes and tells you exactly where your digital

present stands in the AI era. That's it for this week's AI Assembled, the top AI news for

small business. If you found this valuable, share it with a fellow business owner. If you want

a hands-on group that you can work with on your specific AI needs, consider joining business

success academy at freshmediaworks.com slash BSA. Until next week, keep innovating and keep it