Social media companies faced major tests of new policies to protect the integrity of the presidential election into Wednesday, as President Trump, his campaign and others made premature declarations of victory before both national and state races had been called.

Twitter and Facebook have both said they will sanction those who try to declare early victory, a scenario the companies and other tech giants prepared for when considering threats to democracy in the lead-up to the election.

Early Wednesday morning, President Trump posted on both Twitter and Facebook: “We are up BIG, but they are trying to STEAL the Election. We will never let them do it. Votes cannot be cast after the Polls are closed!”

His post was hidden with a warning on Twitter almost immediately. It remained unlabeled longer on Facebook; the company added a label that read, “Final results may be different from initial vote counts, as ballot counting will continue for days or weeks.”

Shortly after, the president held a news conference his campaign live-streamed on Twitter in which he said “we did win this election” and declared other state victories as ballots in many jurisdictions were still being counted. His Twitter account and his campaign’s did not immediately repeat those claims. Still, several pro-Trump partisan outlets with large followings wrote articles that repeated his words without context, while mainstream outlets, including the Associated Press, also tweeted the claims in quotes.

A video of the eight-plus minute speech posted to Trump’s Facebook page received a label that read, “Final results may be different from initial vote counts, as ballot counting will continue for days or weeks after pools close.” Twitter said that video of the news conference is not a violation of its policy, and reporting on a news conference is also allowed.

Twitter also applied a label to a Trump campaign tweet that claimed early victory in South Carolina without linking to an official news source, as well as one by Florida Gov. Ron DeSantis (R) asserting victory for the president in his state. The tweets violated Twitter’s policies against calling a victory before two out of seven approved news outlets have done so. The labels said that “official sources may not have called the race when this was tweeted” and provided a link to authoritative results.

Identical Facebook posts by Trump’s campaign on South Carolina and by Gov. DeSantis only had the usual election updates link, which was the same generic information label Facebook had affixed to most election-related content all day.

Trump allies, including some statewide elected leaders, early Wednesday morning intensified their accusations online that some news networks were selectively calling certain races to damage the president. Some suggested baselessly to their hundreds of thousands of followers on social media that Democrats would use the extra time to engineer more votes. Trump’s campaign sent a similar email to supporters, and the president reiterated those claims during his early-morning speech.

The developments early Wednesday followed a fresh barrage of election-related misinformation targeting voters, the latest in a voting period that has been marred by misleading narratives across social media.

#StopTheSteal, a hashtag associated with alleged voter fraud and a Democratic theft of the election, was shared more than 50,000 times. It was driven largely by right-leaning influencers, including Donald Trump Jr. and Ann Coulter, amplifying isolated incidents that were then taken out of context, according to researchers. One video, in which a pro-Trump poll watcher was mistakenly prevented from entering a Philadelphia polling location, racked up more than 287 million likes, retweets and views across Twitter by the afternoon, where it was framed in some cases as evidence of efforts to steal the election, according to researchers.

Twitter penalized several high-profile accounts, including one from a Trump campaign official and a network connected to former Trump adviser Stephen K. Bannon. Twitter also removed a post, shared tens of thousands of times from a screenshot on Instagram, in which a person falsely claiming to be a poll worker in Erie, Pa., said he had thrown out hundreds of Trump ballots.

Late Monday, in a tweet that Twitter restricted with a label, Trump tweeted that the Supreme Court’s recent decision about Pennsylvania mail-in ballots will “induce violence in the streets.” He added, “Something must be done!”

The narratives that spread across social media contrasted with what many observers described as a largely peaceful day of voting nationwide. Many of the attempts appeared specifically targeted at voters in swing states, particularly in the battleground state of Pennsylvania. Some of the messaging, like the president’s, intimated that violence could take place. Trump’s statements echoed concerns among elected officials and businesses, which boarded up storefronts ahead of Election Day.

“My biggest fear is the potential for physical violence that we didn’t have in 2016,” said Alex Stamos, head of the Stanford Internet Observatory and a former Facebook chief security officer, said on a media call Tuesday morning from the Election Integrity Partnership, a coalition of misinformation researchers that examined the #StopTheSteal hashtag.

The lead-up to the 2020 election has been uniquely influenced by social media, particularly because in-person campaigning has been more limited by the global pandemic. Both President Trump and Democratic candidate Joe Biden’s campaigns have spent millions of dollars on social media and other targeted advertising in recent weeks.

But researchers have cautioned that domestic disinformation has also taken on increased force in this election cycle, as high-profile influencers and problematic groups attempt to spread lies online and even the president uses his Twitter account to share misinformation to his more than 87 million followers.

Facebook, Twitter, Google and Google-owned YouTube collectively held more than 100 scenario-planning exercises, launched new policies including prohibitions on premature declarations of victory and calls to violence, and taken unprecedented enforcement actions, according to the companies.

They also have formulated detailed plans on how they will flag whether the election has been decided, partnering with media outlets to try to slow the spread of misinformation. Facebook and Google have banned political and social ads with the close of polls Tuesday, and Twitter has banned them entirely.

They are trying to prevent a repeat of 2016, when in the weeks after the election, they discovered that their platforms had been abused by Russian operatives who successfully sowed disinformation among American voters. Russian state media and related accounts have been active in the 2020 election, including pushing misinformation about voter fraud on Election Day, researchers say, but their influence has so far appeared smaller than the effect of homegrown disinformation.

Some deceptive efforts seemed to blur that distinction. Twitter on Tuesday removed a network of accounts pushing anti-Biden conspiracy theories and other extremist content as part of a coordinated onslaught attributed by Alethea Group, a consultancy tracking disinformation, to Chinese billionaire Guo Wengui “with support from Stephen K. Bannon.” The men in recent weeks have sought to leverage their combined business and media influence to train attention on unverified claims about the Biden family.

The online activity, which spanned numerous platforms, boosted pro-Trump talking points and spread baseless or misleading accusations about the former vice president and his son Hunter, as well as about urban unrest and other hot-button issues. It also spread narratives critical of the Chinese Communist Party, all the while seeking to drive clicks to news sites run by Guo Media, Alethea Group found.

Twitter said it had suspended about 150 accounts for violating the company’s rules against platform manipulation and spam, specifically coordination geared toward “abusive behavior.” Bannon did not respond to a request for comment.

Throughout Election Day, Twitter labeled some posts “disputed” and potentially “misleading about an election or other civic process,” including several #StopTheSteal posts that suggested that fraud was rampant. But many of them remained on the site unflagged.

Late in the afternoon, YouTube also took down videos that purported to show Election Day violence but were actually rehashing of video clips from previous protests, the company said. The Election Integrity Partnership said that it reported to the video app site TikTok certain accounts that were claiming there would be violence if Biden won. TikTok said it blocked one of the videos.

And early Wednesday, Facebook also added a label warning that votes were still being counted and that the election had not yet been called to a post by President Trump that read, “I will be making a statement tonight. A big WIN!” On Twitter, the same post stayed up with no label 45 minutes after it was posted.

Facebook confirmed it had applied the labels to the two Trump posts early Wednesday. The same “votes are being counted” warning also then appeared on all recent Trump Facebook posts and posts from Joe Biden’s official account, even if they didn’t claim victory.

On Tuesday, officials in Erie County, Pa., disputed the claims in the viral post regarding Trump ballots being discarded. “The person making the statements does not work in any way with Erie County,” the county said via its Twitter account.

The dissemination of misleading narratives was sometimes coordinated behind the scenes in an effort to create the appearance of an online groundswell. A post on 8kun, an anonymous image board at the center of the pro-Trump QAnon conspiracy theory, advised the use of particular hashtags, from #Watchyourballot to #VoteInPerson to #Trump2020Landslide.

The president’s tweet Monday night about violence in Pennsylvania was labeled by Twitter with a notice that voting by mail and voting in person have a long history of trustworthiness, and that voter fraud is “extremely rare.” It also took actions to restrict the spread of the tweet. But the tweet had already been retweeted more than 55,000 times before the social media company throttled it, according to the Election Integrity Partnership.

Facebook appended a label to the same post on its site about the security of mail ballots. Still, it received internal pushback from its own employees saying the company should do more, according to internal communications viewed by The Washington Post.

The light touch from the world’s largest social network alarmed David Brody, counsel and senior fellow for privacy and technology at the Lawyers’ Committee for Civil Rights Under Law.

“It’s really important for the platforms to raise up the authoritative sources and algorithmically downlink conspiracy theories and unsubstantiated rumors,” Brody said, warning about the possibility of the president’s words leading to “real-world violence.”

The #StopTheSteal hashtag gained momentum Tuesday as users and right-leaning influencers spread the banned poll watcher video and other isolated incidents of improper practices or glitches at polling locations, according to First Draft News, a nonprofit that focuses on tackling misinformation. Pro-Trump users popularized the #StopTheSteal hashtag during the 2018 midterm elections as part of similarly baseless allegations of wide-scale voter fraud. There were also some signs that the hashtag had been promoted by bots.

Zignal Labs, a media intelligence firm, said the hashtag went from just a few dozen mentions at 8 a.m. Tuesday morning to more than 2,000 every 15 minutes by 8:15.

The video of a Trump poll worker wrongly being denied entry to a polling place in Philadelphia went viral on Twitter with that hashtag and commentary around efforts to steal the election. A local polling judge incorrectly told him that his certificate was acceptable at only one location in the city, when in fact it applied at any.

Kevin Feeley, a spokesman for Philadelphia City Commissioner Lisa Deeley, said that the location’s judge of election made an “honest mistake” in preventing the watcher from entering the location and that the commissioner’s office acted quickly in informing him of the correct rules.

The poll watcher did not reenter that particular location, but Feeley said that he did gain admittance to another polling location in Philadelphia.

Twitter also sanctioned tweets by Trump campaign official Mike Roman and the Philadelphia Republican Party for making misleading claims about the voting process in Pennsylvania. The tweets made misleading claims about reports of people submitting multiple mail-in ballots, and putting up partisan signage at polling places.

Complaints leveled on Twitter by the Philadelphia GOP and by members of the Trump campaign did not yield formal reports to authorities, said Jane Roh, a spokeswoman for the Philadelphia district attorney’s office, “which arguably raises questions about the actual intent of these posts.” She added, “Where we have been able to identify a location, our election task force has responded and reviewed, and determined the allegations are unfounded.”

Narratives pushing unproved allegations of widespread voter fraud have been circulating on social media for months, including from Trump, his adult sons and sympathetic media outlets. Stories have been taken out of context, such as a claim that ballots that were found in a ditch in Wisconsin were put there on purpose to hurt President Trump.

A video clip of Biden that was deceptively edited to make it appear that he was admitting to voter fraud racked up more than 17 million views over the past week, according to the left-leaning human rights group Avaaz.

Disinformation experts said social platforms were acting nimbly to combat posts intimating violence and misinformation but were not taking action to punish repeat offenders.

“You see the same people over and over again,” Stamos said on the call. He suggested that during certain highly sensitive periods, such as elections, the companies should consider enacting stiffer penalties such as freezing people’s accounts after fewer violations.

Craig Timberg, Cat Zakrzewski, Rachel Lerman, Drew Harwell and Tony Romm contributed reporting.

correction

An earlier version of this story incorrectly said Twitter labeled a Trump campaign tweet on Florida.