> daily_signal(2026_07_03)

Getty and Shutterstock called off their $3.7 billion merger because AI reset the price of an image library, while X asked the FTC to lift the privacy order it earned.

PickBits Daily Signal · Friday, July 3, 2026

By Mark Pickering · 9 min read · July 3, 2026

// tl;dr

Getty and Shutterstock abandoned their $3.7 billion merger, and the regulator was the trigger, but the real reason is AI. UK regulators would only clear the deal if Shutterstock sold its editorial business, which Getty's board rejected outright. Days before pulling out, Getty signed a deal to license its whole library to OpenAI. The market repriced: value now sits in who you can license images to, not in owning more of them.
X asked the FTC to throw out the 2022 privacy order it earned as Twitter, arguing the order slows its AI. EFF and allies filed to reject that. The order stemmed from the misuse of 140 million users' security data for ad targeting; it runs through 2042. EFF's answer: renaming the company doesn't erase the obligation; the compliance cost is trivial relative to a ~$200 billion valuation; and the AI era raises the privacy stakes rather than lowering them.
A survey of more than 13,000 Georgia teachers found 59% now use AI to teach and 95% use it to plan lessons, even as a majority warn it is hurting how students learn. Nearly 90% call it a positive for their own classroom. But management is improvised: only about half have defined AI cheating, and 43% use detection software widely known to be unreliable.
California put Claude on every state desk at half price, aimed at DMV wait times, Medicaid workflows, and cyber defense. The deal covers all state agencies plus cities and counties, with free training, and reaches roughly 39 million residents. Governor Newsom's stated condition: AI augments public workers; it does not replace them.

The thread under today is a single question asked in four rooms: once AI is the thing setting the price, dodging the regulator, sitting on the desk, and reporting for duty, who is it actually answering to? Getty and Shutterstock spent eighteen months trying to merge and this week walked away, not because a regulator finally won, but because AI quietly changed what a giant image library is even for. X went to the FTC and argued the privacy order it earned by misusing 140 million people's data should be lifted, because that order is a drag on its AI. Georgia counted its teachers and found that AI has already won the teacher's desk, while the same teachers warn that it is hurting the kids. And California handed Claude to every agency it runs, on the condition it helps the workers rather than replacing them. The market, the regulator, the classroom, the public counter. One question in each.

Getty walked from its merger and sold to OpenAI, X asked the FTC to drop its privacy order, Georgia's teachers went all in on AI while warning about their students, and California put Claude on every state desk.

1. Getty and Shutterstock called off their $3.7 billion merger, and the real reason is AI.

Buying your biggest rival made sense right up until a model could make the product for free.

Getty Images and Shutterstock have been trying since January 2025 to combine into a single $3.7 billion company that would have held the two largest stock-photo libraries on earth, with Shutterstock alone listing around 450 million images and the pair projecting $150 to $200 million in savings. On June 30 they called it off. The trigger was the UK's Competition and Markets Authority, which said it would only clear the deal if Shutterstock sold off its editorial business, the archive of real news and event photography that newsrooms and publishers rely on. Getty's board treated that as a non-starter, voted unanimously against it, and moved to terminate the agreement, with a reported July 6 deadline. But the regulator is not the real story. Days before pulling out, Getty signed a deal to license its entire library to OpenAI, feeding it into ChatGPT.

Two years ago the logic was obvious: merge, get bigger, own more of the catalog, and set the price by scale. AI broke that logic. When a model can generate a passable image for nothing, owning fifty million more generic ones adds little, and what suddenly matters is whether you can license the images you already hold to the companies training those models. Getty read the room and switched teams, from consolidating against Shutterstock to selling to OpenAI. We flagged this turn when Getty first started signing AI-licensing deals, and the collapsed merger is that same move finishing. The catch is that Getty is now playing both sides of a fight it helped start: it has sued AI image generators for training on its pictures without permission, and it is selling those same pictures to one. What a library is worth now has less to do with how big it is than with which AI company you are willing to sell it to.

Hollywood Reporter June 30 2026 Getty Images and Shutterstock terminate their $3.7 billion merger first announced January 2025 after UK Competition and Markets Authority required Shutterstock to divest its editorial business Getty board voted unanimously against and instead pivoted to license its image library to OpenAI through ChatGPT — hollywoodreporter.com · June 30, 2026

Why this matters: If you make anything for a living, photos, writing, music, code, the Getty story is the template for what happens to your work once a model can imitate it. The market splits into cheap machine-made versions and a smaller premium tier for the licensed, provenance-backed real thing, and the money shifts from selling to humans toward licensing to the machines. Action this week: If you hold rights to any creative work, check whether your contracts or terms even mention AI training, because most were written before it existed, and decide in writing whether your work is licensed for model training and at what price, kept separate from ordinary use. If you buy visual content for a business, sort your uses into the ones that legally need a real, rights-cleared, indemnified image, anything with a real person, a news claim, or a regulated product, and the ones where AI-generated is fine, before you ship a generated image into a slot that needs provenance and legal cover.

hollywoodreporter.com: Getty Images and Shutterstock terminate $3.7 billion merger after UK regulator demanded editorial divestment (June 30, 2026)
petapixel.com: Getty plans to terminate Shutterstock merger because of UK regulations (July 1, 2026)

2. X wants out of the FTC privacy order it earned, and its argument is AI.

X's whole case is that its AI matters too much to keep the old privacy leash on.

On May 15, X Corp asked the Federal Trade Commission to throw out or soften the consent order it has lived under since 2022, the one it earned as Twitter by quietly repurposing the phone numbers and email addresses that 140 million users had handed over for account security and using them to target ads. That drew a $150 million fine and reporting obligations that run all the way to 2042. X's argument now is that the reporting is burdensome and, more pointedly, that it hobbles the company's AI leadership. On July 2, the Electronic Frontier Foundation and allied groups filed to say the petition should be rejected.

X is making a specific bet: that "we need this for AI" has become a strong enough phrase to pry open a privacy order. EFF's answer is the one worth hearing. Changing the name from Twitter to X does not erase obligations that bind the same company. The compliance cost X calls a burden is a rounding error against a roughly $200 billion valuation. And the AI era is the reason to keep the order, not drop it, because X already wired Grok up to user data without anyone meaningfully agreeing to it, took a data breach in 2025, and now carries a newer risk that did not exist when the order was written, that a cleverly written prompt can pull training data back out of a model. We have been tracking this exact pattern for weeks, privacy invasions built into the AI pipeline by design, from Amnesty's reports to the data brokers Senator Wyden keeps prying loose. The tell is simple. The moment a company says its AI is too important to be watched is the moment the watching matters most.

Electronic Frontier Foundation July 2 2026 EFF and allies file that X Corp petition to the FTC to waive its 2022 privacy consent order should be rejected the order stems from Twitter misleading 140 million users by repurposing security data for ad targeting a $150 million fine reporting runs to 2042 X argues it hinders AI leadership EFF says AI era raises the stakes — eff.org · July 2, 2026

Why this matters: If you have ever handed a platform your phone number just to secure your account, this is about whether that number can quietly become ad-targeting fuel, or training data, later, and whether a regulator can still do anything about it. Action this week: On X specifically, assume the data you gave for security and anything you typed into Grok has been usable for training and targeting, and that a 2025 breach may already have exposed some of it, so turn off personalization and data-sharing where the settings exist and pull any recovery phone or email you would not want repurposed. If you own privacy or legal at a company training models on user data, read the filing as the regulatory weather: "we need it for AI" will not get an existing consent decree lifted, so map now whether any data you collected for one stated purpose is quietly feeding a model, before a regulator or a plaintiff maps it for you.

eff.org: EFF and allies say X's FTC petition to waive its privacy-violation order should be rejected (July 2, 2026)

3. 13,000 Georgia teachers now use AI to plan class, and the same teachers say it is hurting their students.

The teachers who use it every day are the same ones raising a hand about the kids.

Georgia just produced one of the largest real datasets we have on AI in a classroom: a survey from the state's Department of Audits and Accounts, published June 26, drawing more than 13,000 teacher responses. The headline number is that 59% of them now use AI for teaching, and 95% use it to plan and prep lessons at least occasionally, more than half of them weekly. Nearly 90% of the teachers who used it said it helped their classrooms, citing time saved, better materials, and more student participation. So far this reads like an adoption success story. Then the same teachers turned around and said the opposite about their students.

The split is the whole story. On the teacher's side of the desk, AI has already won, quietly, without a policy, a year before most districts will write one. On the student's side, the people closest to the kids are the alarm. Majorities of middle- and high-school teachers called AI's effect on student learning negative, and high-school teachers reported the highest student use, a majority saying kids lean on it for at least half their assignments. The management is improvised: more than half now require in-class writing to blunt it, only about half have even defined what counts as AI cheating, and 43% run detection software that is famously unreliable. We have watched other systems reach for the cleaner rule, like Norway, which drew a hard line that under-17s use AI only under supervision, on the logic that a kid has to be able to do the thing before a machine does it for them. Georgia is the same problem with the line not yet drawn. Superintendent Richard Woods framed it as "AI should always be a tool, never a replacement," which is the right sentence. The survey is what it looks like before anyone has enforced it.

Georgia Public Broadcasting June 26 2026 survey by the Georgia Department of Audits and Accounts of more than 13000 teachers found 59 percent use AI for teaching 95 percent for lesson planning nearly 90 percent call it positive for their classroom but majorities of middle and high school teachers say it is negative for student learning 43 percent use detection software — gpb.org · June 26, 2026

Why this matters: If you have a kid in school, AI is already in both halves of their day, in how the lesson gets built and, by high school, in how their classmates do the homework, whether or not the school will say so. Action this week: Ask your kid's school one plain question, what is the actual AI policy, how is student use defined, and how is it checked, and if the answer is a shrug, that is the finding. Push for assignments that build the skills a chatbot can't shortcut, in-class writing, reading aloud, showing the work. If you run a school or district, treat 59% of 13,000 teachers as proof it is already in your building, and write the policy that sanctions the prep uses teachers love, sets grade-specific rules for students with a shared definition of cheating, and puts the money into assessment design instead of detectors that don't work.

gpb.org: More than half of Georgia teachers now use artificial intelligence to prepare for class (June 26, 2026)
the74million.org: More than half of Georgia teachers now use artificial intelligence to prepare for class (June 2026)

4. California just put Claude on every state desk at half price, aimed at the lines people actually wait in.

A genuinely good use of AI, riding on one promise: a human stays on the calls that decide your case.

On June 29, Governor Gavin Newsom and Anthropic announced what both call a first-of-its-kind deal: every California state agency, plus any city or county that wants in, gets Claude at a 50% discount through the state's shared IT portal, bundled with free training and support from Anthropic's own engineers. And it is not abstract. The DMV is using it to cut customer wait times; the Department of Health Care Services is integrating it into the workflows behind Medicaid; the technology and emergency services departments are partnering on cyber defense; and it supports Engaged California, a platform for bringing ordinary residents into policy decisions. This reaches the most populous state in the country, roughly 39 million people.

This is the constructive one, and it is genuinely good. Cheaper, faster public services are a real thing to want, and Newsom drew the line that matters out loud: AI "should not replace the human work of government," only speed it. It lands the same week we were tracking Washington move to gate and restrict frontier AI, so here is a state betting the other way, on deployment with a guardrail rather than a ban. But the guardrail is a promise until someone enforces it. The places this touches, whether you qualify for Medicaid, how your DMV case gets decided, are exactly the places where an automated shortcut becomes a wrong answer with consequences. Point AI at the DMV line and you save people hours. Point it at a benefits determination with no human on the hook, and you have automated a mistake. The augment-not-replace rule only holds if a licensed human actually stays on the eligibility and medical calls, and if the outputs get audited for the errors and bias these systems reliably produce. So the rule itself is right. What I will be watching is whether the agencies hold that line when the wrong call is the faster one.

Governor of California June 29 2026 Newsom announces first-of-its-kind partnership giving all state agencies plus cities and counties access to Anthropic Claude at a 50 percent discount with free workforce training DMV using it to cut wait times Department of Health Care Services for Medicaid workflows CDT and CalOES on cyber-defense reaching 39 million residents with an augment-not-replace guardrail — gov.ca.gov · June 29, 2026

Why this matters: If you deal with any California agency, the DMV, Medi-Cal, a county office, expect faster AI-assisted service soon, and know that you can still ask whether an automated system touched a decision about you and request a human review of a bad one. Action this week: If you work in government IT or benefits anywhere, treat California's deal as a procurement checklist, demand a documented human-in-the-loop for anything touching eligibility, medical, or legal determinations, require audit logs and bias monitoring, and fund the frontline retraining up front so staff supervise the tool instead of deferring to it. If you are a Californian, the state's own framing is your leverage: AI is supposed to speed the work, not be the final word on your case.

gov.ca.gov: Governor Newsom announces a first-of-its-kind partnership providing Anthropic tools to state agencies (June 29, 2026)
techcrunch.com: Anthropic and Gov. Newsom forge deal allowing California government to use Claude at half price (June 29, 2026)

» What to watch this week

Whether Getty's OpenAI licensing deal becomes the template other rights-holders copy, and whether Getty's own lawsuits against AI image generators survive now that it is selling to one. The tension between suing over training and licensing for training is about to become the whole industry's problem, not just Getty's.
Whether the FTC entertains X's petition or rejects it outright, and whether "we need it for AI" starts showing up in other companies' bids to shed old privacy obligations. If it works once, it becomes a playbook.
Whether any Georgia district turns the survey into an actual policy before the new school year, with a shared definition of student AI use rather than more detection software. The data is in; the question is whether anyone will act on the learning-harm warning that the teachers just filed.
Whether other states copy California's Claude deal, and whether anyone publishes the first audit of an AI-assisted benefits or DMV decision. That audit is the moment the augment-not-replace promise gets tested in public.

Tomorrow's signal lands here.