But Instagram is. Its recommendation algorithms learn what keeps users watching, infinite scroll eliminates the natural stopping point and autoplay serves up the next video without waiting to be asked. Meta spent years making its platforms easier to keep using; now it may have to figure out how to make them easier to put down. 

Now Big Social Media is facing its Big Tobacco moment in court as four states—California, Colorado, Kentucky and New Jersey—try to prove that Meta designed Instagram and Facebook to keep young users hooked. Meta says it faces theoretical penalties of $1.4 trillion, but the bigger question for the product itself is what happens if Meta is forced to restrict features such as infinite scroll, autoplay, notifications and recommendation systems: the same machinery critics call addictive is also part of what makes Instagram compelling to use. These social-media cigarettes, in other words, could be getting a harsh new filter.

Larry Magid, a longtime online-safety advocate who has advised Meta on safety issues, thinks the cigarette comparison only goes so far. For years, he preferred a different vice: chocolate. Unlike cigarettes, Magid argues, social media can have real benefits. Young people can use it to maintain friendships, find communities, organize around causes and access information. He saw excessive use as a risk, but not evidence that the product itself was inherently harmful.

“The algorithms change the whole nature of it, and that’s where my chocolate metaphor went awry,” Magid told Fortune. A chocolate bar, after all, doesn’t refill itself.

Infinite scroll and autoplay do something closer to that. Magid likened the experience to a grocery store automatically delivering more chocolate bars whenever it notices the supply is getting low.

“It encourages gluttony,” he said.

Can Meta make Instagram less addictive without making it worse?

The analogy gets at the dilemma facing Meta. The features being challenged in Oakland aren’t obscure corners of Instagram. They help determine what people see, how easily they move from one piece of content to the next and, ultimately, whether they keep scrolling.

But simply ripping out the algorithm isn’t necessarily the answer.

Magid tried that himself. He switched his Facebook feed to chronological order, removing the recommendation engine that decided what he was most likely to want to see and found it boring.

“I actually went back to the algorithm because I actually found it was benefiting me in some ways,” Magid said.

That experience illustrates the line Meta may have to walk. A recommendation system can surface posts users genuinely want to see without necessarily manipulating them into staying. Magid said the better approach would be to make recommendations less aggressive and put more emphasis on a user’s “social graph”: the friends, communities and interests that person has actively chosen.

That would also represent something of a return to social media’s roots.

Magid has worked with Meta on safety issues since 2005, when Facebook had only recently expanded beyond college campuses to high school students. He has watched the platform shift from a network built largely around interactions among friends and classmates into one increasingly shaped by recommendations, influencers and content from strangers.

“What changed with Meta and other companies…is moving from being a truly social network where friends interact with friends, classmates, friends of friends, into being something that’s been driven by algorithms,” he said.

Magid currently serves on Meta’s Safety Advisory Council and Youth Advisory Council, as well as a safety advisory group for Meta Reality Labs. Meta does not currently make financial contributions to his nonprofit, ConnectSafely, though it compensates the organization for participation on advisory councils and content creation.

His proximity to the company has also given him a view into an increasingly difficult problem: The Instagram an adult sees may look nothing like the Instagram served to a teenager. Magid said his own Facebook and Instagram feeds are relatively benign. He sees aviation content, news and the occasional political disagreement. But some of the teenagers and young adults he speaks with describe something considerably darker.

“I am told by minors, by high school kids and young adults, that they have a very different experience than I do,” Magid said. “They are seeing misogyny, they’re seeing homophobia, they’re seeing racism.”

Personalization is what makes both experiences possible.

The same technology that can figure out Magid likes airplanes can learn what captures a teenager’s attention. That doesn’t make recommendation technology inherently harmful, he said, but it raises the stakes of what platforms choose to optimize for and how aggressively they keep serving more of it.

Magid said he would prefer feeds dominated by people users know and interests they have explicitly selected, with fewer recommendations pushed at them simply because the system predicts they will engage.

And he doesn’t think that necessarily has to hurt Meta’s business. If the company is required to dial back some of its recommendation systems or engagement features for younger users, Magid said the result could increase trust without dealing a major blow to Meta’s revenue.

“I personally don’t think that’s going to have a huge impact on their revenue,” he said. “I think, in fact, it might increase their revenue if it creates more trust.”

Meta has told Fortune that less than 1% of its revenue comes from teens on Instagram. The company disputed the states’ allegations, arguing they have not shown that people in their states were harmed by the features at issue and saying it has created strong protections for teens.

But Magid doesn’t want the answer to be kicking teenagers off social media altogether. He still sees access to online communities and expression as valuable for young people. Instead, he wants Meta to build a version of the product that gives teenagers more control over what reaches them and puts less emphasis on keeping them engaged for as long as possible.

In other words, the challenge isn’t making Instagram something teenagers don’t want to use. It’s making an Instagram they can more easily choose to stop using.

Meta’s board can’t unlearn what comes out in court

Changing the product may not be Meta’s only challenge. There is another reason the company may have to respond even if it ultimately prevails in Oakland: Its board can’t unlearn what comes out at trial.

The trial is creating a public record of internal documents, testimony and allegations about what Meta knew about potential harms to young users and how people inside the company responded. For Meta’s board, that information could matter long after a verdict.

Stavros Gadinis, a professor at UC Berkeley School of Law who specializes in corporate governance, said directors generally have substantial protection when they make business decisions after consulting lawyers and advisers. But the more information that surfaces about a serious corporate risk, the harder it becomes for a board to argue it had no reason to intervene.

“The plausible deniability that they were able to, let’s say, defend up to this point retreats a little bit,” Gadinis told Fortune. “It becomes harder and harder and harder to defend as more and more evidence surfaces.”

Gadinis emphasized that the legal bar for holding directors responsible for failing to oversee corporate risks is high. The Oakland case does not automatically create liability for Meta’s board, and even a victory for the states would not mean directors themselves violated the law.

But a loss could make the risks confronting the board going forward harder to ignore.

“If Meta lost the trial, and the court found that indeed the plaintiffs are right and the effects are as they described, that would definitely be a red flag,” Gadinis said. “That would definitely put the board on notice that you cannot keep behaving like that.”

Meta is already facing such a warning from New Mexico, where a jury earlier this year found the company liable for 75,000 violations of the state’s consumer protection law in a separate child-safety case. Meta is appealing. Even a victory in Oakland would require looking at why Meta won, Gadinis said. A decision rejecting the states’ underlying theory would send a different signal than a victory because prosecutors failed to prove one element of their case.

For directors, the distinction matters because the next legal question could be less about what Meta did before these trials and more about what it does after them.

“What’s done is done,” Gadinis said. “They cannot change what they’ve done in the past, but they can always change what they’re doing in the future.”

That puts Meta’s product and its board in versions of the same predicament. The company is learning more about the potential risks created by systems built to maximize engagement, while courts are considering whether those systems cross a legal line. Ignoring that information becomes harder each time another case puts it on the record.