For years, the debate over teenagers and social media has largely revolved around one question: how much responsibility should technology companies take for what happens to young users on their platforms?

Meta is now facing that question in a much more concrete way.

The company behind Instagram and Facebook has agreed to a sweeping set of changes affecting users under 18 in the United States, following an $18 billion settlement with 48 states, the District of Columbia and U.S. territories.

The agreement goes well beyond asking parents to monitor what their children do online.

Meta has committed to stronger age checks, limits on how long teenagers can use its platforms, overnight restrictions and changes to some of the features available to younger users.

Most of the measures are expected to remain in place for 10 years.

The settlement is significant, but the more interesting story is what it says about the future of social media.

For years, platforms were built around keeping people engaged for as long as possible. Now, at least for younger users, Meta is being pushed toward designing its products around something very different: knowing when to stop.

A Two-Hour Limit for Users Under 18

One of the most visible changes is a daily time limit.

Under the agreement, users under 18 in the United States will face a two-hour daily limit on Meta’s platforms. The restriction will be enabled by default and can only be disabled with parental permission.

That is a substantial change for a company whose business depends heavily on people spending time inside its apps.

Instagram in particular is designed around an endless stream of photos, videos and recommendations. There is no natural stopping point. A user can finish one video and immediately receive another.

The new rules introduce an artificial stopping point.

Whether teenagers actually spend less time on the platforms will depend partly on how the system is implemented and how easy it is to work around. But the principle itself is important.

For perhaps the first time, one of the world’s largest social media companies is agreeing to a hard limit on how long younger users can remain inside its products.

Instagram and Facebook Will Shut Down Overnight

Meta is also introducing an overnight restriction for users under 18.

The platforms will be blocked for younger users between midnight and 6 a.m.

The purpose is straightforward: reduce the amount of time teenagers spend scrolling during hours when they should presumably be sleeping.

The measure also recognizes something parents have complained about for years.

Social media does not necessarily stop being available just because it is late at night.

A teenager lying in bed with a phone can continue watching videos, checking messages and scrolling through recommendations long after everyone else in the household has gone to sleep.

The overnight cutoff changes that equation.

It is also another example of Meta moving some responsibility away from parents and putting controls directly into the platform.

Meta Is Tightening Age Verification

Time limits are only useful if the company knows how old its users are.

That is one of the hardest problems facing social media platforms.

Federal privacy rules generally prohibit children under 13 from creating accounts on services covered by the Children’s Online Privacy Protection Act, commonly known as COPPA. Teenagers between 13 and 17 can use social platforms, but they may require additional protections depending on the service and applicable laws.

The difficulty is obvious: people can simply lie about their age.

Meta has already been using artificial intelligence and other signals to estimate whether users are teenagers or adults. The company is now required to strengthen those systems under the settlement.

The agreement also calls for outside testing and measurement of how effectively Meta identifies users’ ages.

That is an important detail.

It is one thing for a company to say that it has an age-verification system.

It is another to measure how often the system gets it right.

There is also a provision that could have a wider effect. When Meta identifies an account belonging to someone under 13, it can check the ages of that person’s friends as well.

That could help the company identify networks of younger children who may have entered the platform by providing false information about their ages.

The Problem With AI Guessing Your Age

Age verification sounds simple until technology has to do it at the scale of billions of users.

A system might examine information about how an account is being used, the language appearing in posts, account behavior and other signals to estimate someone’s age.

But an estimate is not the same thing as proof.

A 17-year-old could look like an adult to an automated system. An adult could also trigger a system designed to identify teenagers.

That creates a delicate balance.

If Meta’s systems are too weak, children can slip through.

If they are too aggressive, legitimate adult users could find themselves incorrectly classified as minors and subjected to restrictions.

That is why independent testing and measurable performance targets matter.

The question should not simply be whether Meta has deployed an AI age-estimation system.

The better question is how accurately does it work, and what happens when it gets someone’s age wrong?

More Safety Features Will Be Turned On Automatically

Another major change involves default settings.

Some of Meta’s safety tools have historically required users or parents to activate them.

The new settlement changes that approach for many protections affecting teenagers.

The two-hour time limit and overnight restrictions, for example, will be enabled by default.

This may sound like a small technical detail, but it has major implications.

Most people do not change the default settings on their applications.

That means a safety feature that is turned on automatically can reach millions of users, while a feature hidden inside a settings menu may be used by only a fraction of them.

This is why critics have repeatedly argued that platforms should make safer choices the default rather than expecting parents and teenagers to find and activate every protection themselves.

What About Instagram’s Recommendation Algorithms?

This is where some of Meta’s critics believe the settlement does not go far enough.

Modern social media is not simply a collection of posts uploaded by friends.

Algorithms decide what users see.

A teenager can open Instagram without following a particular account and still be shown its content because Meta’s recommendation systems believe the material is likely to keep that person watching.

That system is central to the way modern social media works.

It is also at the heart of some of the criticism surrounding teenage use of these platforms.

Former Meta engineering director Arturo Béjar, who testified during the litigation, has argued that simply limiting the amount of time a young person spends on Instagram does not necessarily address the content that keeps them coming back.

That distinction is worth understanding.

Imagine a teenager is exposed to harmful or highly engaging material for two hours.

Reducing that person’s usage to two hours does not necessarily change what appeared in those two hours.

It changes the duration, not necessarily the experience.

That is why some child-safety advocates want greater restrictions on recommendation systems themselves.

Critics Want More Than Parental Controls

The settlement includes several controls that parents can use to manage their children’s experience.

But some advocates argue that this still puts too much responsibility on families.

Josh Golin, executive director of the child-safety organization Fairplay, has criticized the continued role of recommendation systems in directing young users toward potentially harmful material.

His argument gets to a larger issue in the social media debate.

Parents can monitor screen time.

They can restrict access.

They can take away a phone.

But parents cannot see every recommendation algorithm operating behind the screen.

That part belongs to the platform.

If a company knows that a particular feature encourages unhealthy or dangerous behavior, critics argue, the solution should not always be another parental control.

Sometimes the product itself may need to change.

The “Take a Break” Problem

Meta has already experimented with features designed to encourage teenagers to stop scrolling.

One example is Instagram’s “Take a Break” feature, which prompts users to step away after spending a certain amount of time on the platform.

The challenge is getting people to use it.

A notification can be dismissed with a swipe.

That illustrates one of the difficulties with voluntary safety features. A platform can make a tool available without guaranteeing that anyone will use it.

Meta has increasingly moved some protections into its teen accounts by default, including features intended to encourage breaks.

The settlement pushes that philosophy further.

Instead of simply offering young users safety tools, Meta will be required to make some of them part of the default experience.

What Happens to “Likes”?

Meta will also hide like counts for younger users.

This is another interesting change because visible engagement metrics are deeply embedded in social media culture.

Likes, views, shares and follower counts can become measures of popularity.

For teenagers, those numbers can carry an emotional weight that is difficult for adults to appreciate.

Removing visible like counts does not eliminate social comparison. Teenagers can still see content, communicate with friends and judge their popularity in other ways.

But it removes one particularly obvious numerical scoreboard from the experience.

It is another example of Meta changing the design of its platform rather than simply telling young users to behave differently.

Some Features Will Still Be Optional

The settlement does not turn Instagram and Facebook into completely restricted environments for teenagers.

Some protections will remain optional.

For example, users can choose features such as disabling video autoplay or viewing posts chronologically rather than relying on algorithmic recommendations.

That distinction is likely to remain part of the debate.

If a feature can potentially reduce exposure to unwanted or harmful content, should teenagers have to turn it on themselves?

Or should the safer setting be the default?

There is no simple technical answer to that question.

There is, however, a growing recognition that product design influences behavior.

The way an application is built can determine whether people stop scrolling, continue scrolling, click on something or return to the platform later.

Why the Settlement Matters Beyond Meta

The changes apply to Meta’s U.S. services, but the consequences could extend much further.

Technology companies rarely build completely different products for every market if they can avoid doing so.

If new safety systems work well in the United States, there could be pressure to introduce similar protections elsewhere.

Governments in Europe, Africa and other regions are already debating how social media companies should protect children online.

The issue is therefore much larger than Instagram and Facebook.

It concerns the future design of digital services used by young people around the world.

Meta Still Has Something to Prove

The settlement creates new requirements, but it does not automatically prove that Meta’s platforms have become safe for teenagers.

That will take time.

Researchers and child-safety advocates will be watching whether the new restrictions actually change behavior, whether age-estimation systems work as promised and whether harmful content becomes less accessible to young users.

They will also be watching the recommendation algorithms.

That may be the hardest issue of all.

A social network can limit the amount of time a teenager spends online while leaving the underlying content-ranking system largely intact.

If the most engaging material is also the most harmful, critics will argue that the deeper problem has not been solved.

What Parents and Technology Buyers Should Watch

For parents, the immediate lesson is not that Instagram and Facebook have suddenly become safe.

The new controls should make it easier to manage certain aspects of a teenager’s online experience, but parents will still need to pay attention to what their children are seeing and how they are using social media.

For technology companies, however, the implications are even broader.

Meta’s settlement could accelerate a shift in how digital products are designed.

For years, technology companies have optimized applications around engagement. More time on the platform generally means more opportunities to show advertising and collect signals about users.

The new rules introduce another consideration: well-being.

That could eventually influence everything from recommendation engines and notifications to default settings and age verification.

The Bigger Change Is in the Design Philosophy

The most interesting part of Meta’s new rules may not be the two-hour limit or the midnight cutoff on their own.

It is the underlying change in philosophy.

Social media platforms have traditionally asked users to control themselves.

Want to spend less time scrolling? Set a reminder.

Don’t want to see certain material? Change your settings.

Want more privacy? Find the appropriate option.

The new approach puts more responsibility on the platform itself.

For younger users, some protections will be automatic.

That does not solve every problem associated with social media. It may not even solve the problems critics consider most serious.

But it represents an important shift.

The technology industry is gradually being pushed to recognize that the way a platform is designed can be just as important as the rules written around it.

For Meta, the next challenge will be proving that these changes work in practice.

For parents, regulators and other technology companies, the more important question may be what happens next.

Artificial intelligence is no longer confined to chatbots that answer questions, write emails or summarize documents.

Google wants it doing much more inside some of the world’s most tightly controlled businesses.

The company is taking its Gemini AI into banking and legal work with two specialized enterprise offerings: Gemini Enterprise for Financial Services and Gemini Enterprise for Legal.

The move is worth watching because banks and law firms are not ordinary software customers. They deal with confidential information, strict regulations and decisions where an AI mistake can have consequences far beyond an embarrassing chatbot response.

Google’s answer is to build AI around the way these industries actually work.

That means connecting Gemini to professional databases, internal company information and software that employees already use, while putting security and access controls around the system.

In other words, Google is trying to make Gemini less of a chatbot and more of a workplace tool.

Why Google Is Going After Banks and Law Firms

There is a reason these two industries are interesting targets for enterprise AI.

Both generate enormous amounts of information.

A bank may have analysts studying company filings, market movements, credit information and financial statements. A law firm can have thousands of pages of contracts, court documents, correspondence and regulatory material sitting in its systems.

Finding the right piece of information can take hours.

The work itself is often repetitive, but the consequences of getting it wrong are not.

That creates an opportunity for AI. It can search, organize, compare and summarize information much faster than a person working through documents one by one.

But speed is only half the equation.

A bank cannot simply upload sensitive customer information into an AI system without knowing how that information will be handled. A law firm has an even more obvious concern when its files contain confidential client material.

For these businesses, the question is not just whether AI is clever.

Can it be trusted with the company’s information?

That is the market Google is trying to address.

Gemini for Financial Services

Google’s Gemini Enterprise for Financial Services is built around the research and analytical work carried out by financial professionals.

The platform includes specialized AI skills and can connect with financial data services used across the industry.

Google lists providers including FactSet, LSEG, Moody’s, MSCI, PitchBook and S&P Global. It can also work with information from the U.S. Securities and Exchange Commission’s EDGAR database.

That changes the nature of what a financial AI assistant can do.

Instead of asking a general chatbot a question about a company and hoping its answer is current, an analyst can work with data from sources that are part of the financial research process.

Google has built the system to handle tasks such as company research, financial analysis, peer comparisons, credit research and risk assessment.

For someone working in a bank or investment firm, the attraction is obvious.

Research often involves gathering information from several places before the actual analysis can begin. If AI can handle some of that groundwork, an analyst has more time to concentrate on the part that requires professional judgment.

That is a much more realistic use of AI than the idea that a chatbot will simply replace the analyst.

Financial research is becoming an AI workflow

Consider a simple example.

An analyst is asked to prepare a report on a company. The assignment might require recent financial results, regulatory filings, information about competitors, market data and an assessment of potential risks.

Much of the early work involves searching.

The analyst has to find the documents, compare numbers, check dates and pull together information from different databases.

An AI agent can potentially take on a significant portion of that work.

The analyst still has to decide whether the information makes sense and what it means. But the machine can take care of some of the less interesting work that comes before the decision.

That is where enterprise AI could quietly have its biggest impact.

Not by replacing people overnight, but by removing some of the hours of routine work surrounding them.

Deutsche Bank Has Been Involved

Google did not develop the financial product entirely from a technology laboratory.

Deutsche Bank has worked with Google Cloud on financial AI applications, including research and analysis, giving the technology company access to the kinds of workflows used by a major global bank.

CME Group is also among the financial organizations Google has highlighted in connection with its enterprise AI work.

That industry involvement matters.

Banks tend to be conservative about new technology for good reason. Their systems handle money, identities and highly sensitive information.

An AI product can be technically impressive and still fail to gain traction if it does not fit the way financial institutions operate.

Gemini for Legal Work

The legal offering tackles a similar problem from a different angle.

Gemini Enterprise for Legal is aimed at work such as contract review, due diligence, legal research and regulatory monitoring.

Legal professionals already spend a great deal of time searching through documents. AI is particularly well suited to this kind of information-heavy work.

But there is an important catch.

A legal answer that merely sounds convincing is not good enough.

Lawyers need to know where information came from, whether a legal authority is genuine and whether the material being used is relevant to the matter at hand.

Google has therefore designed the legal offering to work with legal information and the systems used by law firms and corporate legal departments.

The company’s list of integrations includes iManage, NetDocuments, Relativity, Everlaw and Harvey. Google also points to CourtListener as a source of legal information.

These connections may ultimately be more important than the Gemini name itself.

A lawyer does not need another isolated chatbot.

What is more useful is an AI system that can sit alongside the firm’s existing document management, research and case-management tools.

The Security Question Comes First

For both industries, this may be the most important part of the story.

Companies are becoming increasingly interested in AI, but many are also asking uncomfortable questions about what happens to their data once it enters an AI system.

Google’s enterprise offering is designed around those concerns.

The company says customer data remains under the customer’s control and that prompts and outputs are not used to train Google’s foundation models. Gemini Enterprise also includes enterprise security and governance capabilities intended for organizations managing sensitive information.

For a bank, that could mean working with financial research without exposing proprietary information to an outside public system.

For a law firm, it could mean allowing AI to work with internal documents while maintaining the permissions that determine which employees can see particular files.

That second part is easy to overlook.

Imagine a large law firm handling hundreds of cases. A lawyer working on one matter should not suddenly gain access to confidential documents belonging to another client simply because an AI system has access to the firm’s entire database.

The AI has to respect the organization’s existing boundaries.

The same principle applies inside a bank.

An employee’s access to information depends on their role. Enterprise AI therefore needs to understand permissions, not just language.

This Is Where Google Wants Gemini to Go

The bigger story here is not really about two new Gemini products.

It is about where Google thinks enterprise AI is heading.

The consumer version of generative AI made people comfortable with the idea of asking a machine a question and getting an answer.

Businesses need something different.

They want software that can connect to their own information and perform useful tasks within established workflows.

That is why Google is putting increasing emphasis on AI agents.

An agent can potentially do more than answer a prompt. It can retrieve information, use connected tools, analyze material and complete several steps before returning a result to a human employee.

Think of the difference this way.

A chatbot might tell a financial analyst what it knows about a company.

An enterprise agent could potentially gather information about that company from approved sources, compare it with competitors, examine relevant filings and prepare a research package.

The second approach is much closer to having AI embedded in the workplace.

The Human Still Has to Make the Call

There is a temptation whenever a new AI product appears to ask whether it will replace people.

That is probably the wrong question here.

A bank still needs someone who understands financial risk. A law firm still needs lawyers who can interpret the law and advise clients.

AI can help with the work around those decisions.

That distinction is important because professional responsibility does not disappear simply because software was involved.

If an AI system produces an incorrect financial assessment, someone at the institution will still have to answer for the decision.

If an AI system points a lawyer toward an incorrect legal authority, the lawyer cannot simply blame the software.

The most practical model is therefore human plus AI, rather than human versus AI.

There Are Still Plenty of Reasons to Be Careful

The technology has obvious potential, but it is not a magic solution.

AI can make mistakes.

It can misunderstand a document, draw the wrong conclusion or present an answer with more confidence than it deserves. Connecting the system to better sources reduces some risks but does not eliminate them.

There is also the question of cost.

Large organizations often have complicated technology environments. Connecting AI to multiple databases and applications, maintaining those connections and monitoring access can become a substantial undertaking.

Then comes regulation.

Financial institutions and legal organizations operate under rules that do not disappear simply because an AI agent is involved.

That means businesses will have to decide where AI can work independently, where human approval is required and which tasks should remain entirely under human control.

Those decisions may take longer than the technology itself.

What This Means for Companies Buying AI

For businesses considering an AI investment, Google’s approach offers a useful lesson.

The most important question may not be which AI model is smartest?

It may be which AI system can actually work with our business?

A powerful model that cannot securely access the right information may be less useful than a slightly less capable system that fits neatly into an organization’s existing workflow.

That is why integrations, permissions, data governance and auditability are becoming such important parts of enterprise AI.

The chatbot is only one piece of the puzzle.

The infrastructure around it may ultimately matter more.

Google’s Bet Goes Beyond Banking and Law

Google’s move into financial and legal services is part of a much larger competition taking place across the technology industry.

Microsoft, Google and OpenAI are all looking for ways to move AI from the screen where people ask questions into the systems where businesses actually get work done.

That is potentially a much bigger market.

An AI assistant used occasionally by an employee is useful.

An AI system that becomes part of a company’s research, document management, customer service, financial analysis or compliance processes becomes infrastructure.

And infrastructure is much harder to replace.

For Google, banking and law provide a useful proving ground because both industries demand exactly the things enterprise AI will need to deliver: accuracy, security, traceability and control.

The Bottom Line

Google’s expansion of Gemini into banking and legal work is not simply another attempt to make a smarter chatbot.

It is part of a broader shift toward AI that is designed for particular industries and connected to the software and information businesses already depend on.

For banks and law firms, that could mean less time spent searching through mountains of information and more time spent making decisions.

But the technology will have to earn that trust.

For companies considering enterprise AI, that is perhaps the most important takeaway from Google’s announcement. The winning AI system may not be the one that produces the most impressive demonstration. It may be the one that can quietly do useful work inside a business without compromising the information, controls and professional judgment that keep that business running.

That is a much harder problem to solve—and potentially a much more valuable one.