The migrant caravan in Mexico and the attempted mail bombings of major political figures last week have unleashed torrents of false and misleading reports on social media, testing the limits of costly efforts by Silicon Valley to combat disinformation ahead of the 2018 midterm elections.

Despite hiring thousands of employees and investing in teams dedicated to quelling phony information two years after the problem emerged during the 2016 presidential election, the country’s most influential tech companies have struggled to respond.

Facebook, Twitter and Instagram have resisted demands to remove some of the viral conspiracy theories and extremist content — a reflection both of the gravity of the task and of their belief that they should not serve as arbiters of truth.

The attempted pipe-bomb attacks, which targeted former President Barack Obama and others who have been critical of President Donald Trump, were almost immediately characterized in widely shared Facebook and Twitter posts as a conspiracy engineered by Democrats to undermine the conservative cause. Michael Flynn Jr., the son of the president’s former national security adviser, said in tweets to his roughly 98,000 followers that the bombs amounted to a “political stunt.”

Claims that the bombs were a hoax and slurs against one of the bombs’ targets, liberal philanthropist George Soros, also proliferated widely on the Facebook-owned photo-sharing giant Instagram. Social media researcher Jonathan Albright said the Instagram posts amplified conspiracy theories and “some of the worst hate speech, Hillary Clinton memes and violently anti-Semitic messages I’ve seen to date.”

The caravan, a potent symbol of the brewing migrant crisis at the U.S. border, was portrayed by some prominent conservative figures as a violent horde mobilized for invasion, including through the sharing of falsely labeled images showing a bloodied Mexican policeman that was in fact taken elsewhere in the country in 2012.

That image, first posted on Oct. 21, spread virally on Facebook and Twitter, including through a post by Ginni Thomas, a conservative activist who is the wife of Supreme Court Justice Clarence Thomas.

The hoaxes were amplified by accounts known to echo Kremlin propaganda, according to researchers who say the hoaxes are a form of manipulation they have detected repeatedly on controversial topics since the 2016 election. But the largest sources of disinformation on the caravan and the attempted bombings have come from domestic sources, researchers say.

The continued spread of misinformation last week shows how the sites continue to waver on even the most incendiary views related to potentially real-world violence.

“This is an example of where social media companies have a responsibility not to amplify propaganda that is demonstrably false,” Rep. Ro Khanna, a Democratic lawmaker who represents a part of Silicon Valley, said in a statement. “A newspaper or television station would never claim that the pipe bombs are fake, and they wouldn’t give that perspective the time of day. Similarly, social media companies need to have basic third-party verification so they are not allowing false claims to be retweeted or shared.”

The flood of misinformation has infuriated regulators, who have remained vigilant after other spurious users — including agents of the Russian government — stoked social and political unrest online with divisive messages of their own. To that end, lawmakers are especially wary that these hoaxes and conspiracy theories are resonating, and possibly intensifying, with the midterm elections less than two weeks off.

The tech industry has struggled to balance calls for combating disinformation with concerns about protecting free speech, especially at a time when conservatives have blasted Silicon Valley for a supposed pro-liberal bias.

“We have taken action,” Facebook said in a statement Thursday. “We’ve demoted stories rated false by fact-checkers, like content about police brutality by migrants and pipe bombs, and we’re removing content that violates our policies, like hate speech or support for the bombing attempts.”

Instagram, which belongs to Facebook, didn’t respond to requests for comment.

Twitter said it relies on truthful tweets to correct and neutralize false information on its platform, unless messages break its rules, such as threatening violence. “Accounts that deliberately attempt to disrupt the public conversation, including sharing the same content repeatedly or trying to game trending topics, will face enforcement action pursuant with our policies,” a Twitter spokesman said.