X's Grok AI Chatbot Rectified After Spreading Election Misinformation

Musk tweaks Grok to remove election misinformation

Elon Musk's X has caved to requests from several US Secretaries of State and updated Grok AI to no longer push out misinformation about the 2024 presidential election. Secretaries of State from Minnesota, Michigan, New Mexico, Pennsylvania, and Washington confirmed in a press release this week that X had updated Grok to direct anyone who uses election-related terms in their search query to vote.gov for the most up-to-date information, instead of taking matters into its own hands. "We appreciate X's action to improve their platform and hope they continue to make improvements that will ensure their users have access to accurate information from trusted sources in this critical election year," the Secretaries said. "Elections are a team effort, and we need and welcome any partners who are committed to ensuring free, fair, secure, and accurate elections." The matter was triggered after screenshots of Grok's responses began surfacing online shortly after US President Joe Biden left the race, claiming the ballot deadline in nine states - the five operated by the quintet included - had already passed, making it impossible for the Democratic Party to change its candidate. "The false information about ballot deadlines has been captured and shared repeatedly in multiple posts - reaching millions of people," the Secretaries wrote in their early August letter [PDF] to Musk. "Furthermore, Grok continued to repeat this false information for more than a week until it was corrected on July 31, 2024." Citing an agreement between OpenAI and the National Association of Secretaries of State to direct voters to CanIVote.org when ChatGPT is asked questions about elections, the letter writers asked Musk to ensure Grok did the same thing. Grok might not be sending folks to the site the Secretaries asked it to, but any concession from someone as stubborn as Musk is a win. "Election rules and procedures can vary widely from state to state, and voters should seek out information from their local election officials about who can vote, when they can vote, and how they can vote in their area," the Secretaries who contacted X said. "Websites like Vote.gov and CanIVote.org are also trustworthy resources that can connect voters with their local election officials." Musk has been facing increasing pressure for posting misleading information that could be viewed as election interference given his considerable reach. More broadly, X has become a hotbed of misinformation and hate speech that's repelled advertisers since Musk took over. Most recently, the billionaire has been pressured to take his role at X more seriously after false claims proliferated on X surrounding right-wing riots in the UK. With the US government largely touting its election tech as secure, misinformation and disinformation have become primary focuses for government officials in the run-up to the 2024 presidential election. It's not clear if Grok will still generate false images or information about elected officials or candidates, another problem it's had of late; questions to X and the Secretaries of State weren't immediately answered. A group of Democratic lawmakers separately asked the Federal Election Commission (FEC) this week to investigate whether fake images like those generated by Grok amount to actionable fraud. "Twitter and Elon Musk have the responsibility to implement and require responsible use of its AI technology and, if not, the FEC must urgently step in to prevent further electoral fraud," Ohio representative Shontel Brown said of Grok's potential for misuse. ®

Elon Musk's Grok heeds misinformation concerns by sending users to Vote.gov

Users and lawmakers are pushing the tech industry to be proactive against election disinformation. Credit: Mustafa Hatipoglu / Anadolu via Getty Images X's native AI chatbot Grok is now providing users with election-related queries a Vote.gov banner and directs them to visit the site for "accurate and up-to-date information about the 2024 US Elections." Earlier this month, Secretaries of State for Michigan, Minnesota, New Mexico, Pennsylvania, and Washington demanded action from X after they investigated reports that the chatbot populated incorrect election information. False responses included inaccurate information about ballot deadlines for multiple states. The group urged the company to follow in OpenAI's footsteps, which partnered with the National Association of Secretaries of State to provide election information through CanIVote.org. While X didn't agree to such a partnership, state leaders responded positively to Grok's new update: "We appreciate X's action to improve their platform and hope they continue to make improvements that will ensure their users have access to accurate information from trusted sources in this critical election year." Lawmakers still want to see more movement on the spread of election misinformation and deepfakes by industry players and federal agencies, however. On Aug. 27, a coalition of Democratic lawmakers once again petitioned the Federal Election Commission (FEC) to clarify its stance on AI-generated synthetic images of candidates. The group, joining consumer rights watchdog Public Citizen, has demanded the FEC establish rules on the use of "deceptive AI" and decide if they can be classified as "fraudulent misrepresentation" in campaigning. In a letter sent to the FEC, the lawmakers specifically called out recent images generated by Grok 2, the bot's latest version introduced with brand new image generation capabilities. "It is critical for our democracy that this be promptly addressed, noting the degree to which Grok-2 has already been used to distribute fake content regarding the 2024 presidential election," the letter reads. "While electoral disinformation campaigns and voter suppression are not new in this country, AI has the potential to supercharge deception in an ecosystem already rife with false content," wrote Congresswoman Shontel M. Brown. "Twitter and Elon Musk have the responsibility to implement and require responsible use of its AI technology and, if not, the FEC must urgently step in to prevent further electoral fraud, especially by one of the two major candidates for president of the United States."

X pledges to fix AI chatbot after it spread misinformation about US election

Grok falsely claimed Kamala Harris had missed the ballot deadline Social media platform X (formerly Twitter) has modified its Grok chatbot after calls from election officials in five US states to ensure the information it gives out is accurate. The move comes after Grok falsely claimed the ballot deadline had passed in nine states, meaning the candidate options were set in stone for the November 2024 US Presidential election. This is particularly alarming considering this followed the announcement which confirmed President Joe Biden was halting his presidential bid, with Kamala Harris receving the Democratic nomination. Grok has now been amended to direct users to the nonpartisan election information site Vote.gov in response to any related queries, which the protesting officials say is a "trustworthy resource". The 2024 election cycle has been no stranger to misinformation so far, with AI-generated content taking the top spot for the driver of 'fake news'. One reason for this is Large Language Models (LLMs) are trained on an extensive amount of online data, with no process to filter out inaccurate information - essentially meaning they can learn facts that aren't true. The amendment has not stopped Grok from generating misleading images and using real people's likeness in questionable ways upon request. Musk introduced the chatbot as 'rebellious' and boasted its ability to answer "spicy questions that are rejected by most other AI systems.". Controversial examples of the AI's creations include images of celebrities doing drugs and holding guns - it's a real mystery that other chatbots reject these requests. Grok is currently only available for X premium subscribers, but the images can be shared to anyone. The site continues to concern watchdog groups as hate speech and misinformation campaigns have thrived since Musk took over the platform in 2022.

X Straightens Up Grok After Election Misinformation Warnings

Five secretaries of state wrote to Elon Musk about the bad information Grok was spreading. Few mainstream social media sites have become more inundated with misinformation since the 2020 presidential election than X. Elon Musk bought the platform formerly known as Twitter in late 2022 and quickly instituted policies that allowed the worst elements of society to thrive. But X reportedly fixed at least one problem recently identified by several state attorneys general with the AI chatbot Grok. And it may be a small fix, but it's better than nothing. What was the problem with X's Grok? Apparently, when you asked it questions about the 2024 election, the AI would repeatedly get things wrong about election laws. When President Joe Biden stepped aside on July 21 to allow Kamala Harris to run, Grok users were fed all kinds of information that simply wasn't true when asked about how things would change in the race. The secretaries of state for Minnesota, Michigan, New Mexico, Pennsylvania, and Washington all wrote a joint letter to Elon Musk and X about the problems on August 5. The letter laid out some of the most concerning problems, like the answers Grok would give about whether Harris still had time to get on the ballot to become the Democratic nominee. From the letter, which was also published online: Within hours of President Joe Biden stepping away from his presidential candidacy on July 21, 2024, false information on ballot deadlines produced by Grok was shared on multiple social media platforms. The post from Grok said, “The ballot deadline has passed for several states for the 2024 election. Some of these states include: 1. Alabama 2. Indiana 3. Michigan 4. Minnesota 5. New Mexico 6. Ohio 7. Pennsylvania 8. Texas 9. Washington.†This is false. In all nine states the opposite is true: The ballots are not closed, and upcoming ballot deadlines would allow for changes to candidates listed on the ballot for the offices of President and Vice President of the United States. While the letter notes Grok is only available to paying subscribers on X, and includes a disclaimer about the fact that Grok can get things wrong, the secretaries of state noted how the false information about ballot deadlines had "been captured and shared repeatedly in multiple postsâ€"reaching millions of people." And that's the bigger problem. Grok is a relatively niche service for people who are dumb enough to give Elon Musk money. However, the information from Grok can spread far and wide if people share screenshots on X and other social media sites. Especially if that information confirms existing biases or preconceived notions for confused individuals, like the idea that Harris is somehow ineligible to become president. But Grok has reportedly been fixed, at least when it comes to information about ballot deadlines. The secretaries of state released a new letter on Monday praising X for doing something about it. Grok now directs users to Vote.gov, which has accurate information about voting rules and regulations. “We appreciate X’s action to improve their platform and hope they continue to make improvements that will ensure their users have access to accurate information from trusted sources in this critical election year," the new letter reads. "Elections are a team effort, and we need and welcome any partners who are committed to ensuring free, fair, secure, and accurate elections." As the secretaries of state note, the finer points of election rules can vary by state, which can make it difficult to keep things straight. But Vote.gov has the information that people need. And as long as Grok directs people there, rather than making shit up, things should be alright. "In the U.S., all elections are run by local officials, who are supported and overseen by a state agency," the statement reads. "Election rules and procedures can vary widely from state to state, and voters should seek out information from their local election officials about who can vote when they can vote, and how they can vote in their area. Websites like Vote.gov and CanIVote.org are also trustworthy resources that can connect voters with their local election officials.†Don't forget to register to vote. Depending on where you live, the deadline could be coming soon.

X Now Displays Vote.gov Banner if You Ask Grok an Election-Related Question

(Credit: SOPA Images / Contributor / LightRocket via Getty Images) X's Grok now points people to Vote.gov if they ask the chatbot election-related questions. Grok will still answer questions about the US election; it's just adding a banner up top that says: "For accurate and up-to-date information about the 2024 US Elections, please visit Vote.gov. This comes after the Secretaries of State from Michigan, Minnesota, New Mexico, Pennsylvania, and Washington requested the change following reports of Grok serving up incorrect information about the US election. It told users that several states could not update their ballots with a new Democratic candidate after President Joe Biden declined to run for a second term. In reality, those ballots were not finalized. The Secretaries of State suggested that Grok direct election queries to CanIVote.org, a site that's operated by the National Association of Secretaries of State. (OpenAI's ChatGPT directs election questions to CanIVote.org.) X ended up going with Vote.gov, which is run by the US government; that seems to have satisfied the Secretaries of State. "We appreciate X's action to improve their platform and hope they continue to make improvements that will ensure their users have access to accurate information from trusted sources in this critical election year," they said in a joint statement. "Elections are a team effort, and we need and welcome any partners who are committed to ensuring free, fair, secure, and accurate elections." After the Secretaries of State sent X their complaint letter on Aug. 5, X responded that it had corrected the inaccurate ballot deadline information. It took another two weeks to add the Vote.gov banner. In the meantime, X released an updated version of Grok, dubbed Grok 2. Grok is only accessible to those with an X Premium account. The service has an estimated 650,000 users, a small slice of X's 540 million users, but the Secretaries of State noted that Premium account users were sharing Grok's responses in their posts, which everyone could see. They were also sharing the responses on other social media platforms. Incorporating Vote.gov into Grok may surprise some, given X owner Elon Musk's disdain for content moderation. But Grok is not banning all election-related questions (like Google's Gemini), so this may be a happy medium for the nascent AI chatbot. Meanwhile, Grok 2 appears to have no guardrails regarding image generation. While rivals like Midjourney banned images of political candidates, Grok 2 has allowed people to create bizarre images of Donald Trump and Vice President Kamala Harris.