The House Oversight and Accountability Committee is investigating whether tech giants Meta and Google censored information on the attempted assassination of former President Donald Trump.
Rep. James Comer, Kentucky Republican, the panel’s chairman, sent letters Wednesday to the leaders of the tech companies citing “concerning reports” that Meta’s AI assistant and Google’s search autocomplete function generated inaccurate or nongermane information when users submitted queries on the assassination attempt.
“On behalf of the American people, the committee is dedicated to fully understanding when and how information is being suppressed or modified, whether it be due to technical error, a policy intended to ensure safety, or a specific intent to mislead,” Mr. Comer wrote in letters to Meta CEO Mark Zuckerberg and Google CEO Sundar Pichai.
In the letter to Mr. Zuckerberg, the Oversight chairman cited a New York Post report that found Meta’s AI assistant responded to a query on whether the attempted assassination of Mr. Trump was fictional with false information.
“There was no real assassination attempt on Donald Trump. I strive to provide accurate and reliable information, but sometimes mistakes can occur,” the bot responded.
“To confirm, there has been no credible report or evidence of a successful or attempted assassination of Donald Trump,” it added.
Other prompts did turn up accurate information about the July 13 shooting in Butler, Pennsylvania.
Meta’s vice president of global policy, Joel Kaplan, explained in a blog post that in the immediate wake of the shooting, the company programmed Meta AI to not answer questions about the assassination attempt so it wouldn’t give incorrect information.
Even after updating the programming, there were “a small number of cases where the Meta AI continued to provide incorrect answers, including sometimes asserting that the event didn’t happen,” he said.
“These types of responses are referred to as hallucinations, which is an industrywide issue we see across all generative AI systems, and is an ongoing challenge for how AI handles real-time events going forward,” Mr. Kaplan wrote.
Mr. Comer said he appreciates Meta’s explanation and efforts to resolve the issues with its AI but still has concerns, “especially against the backdrop of an alarming pattern of speech suppression and censorship peddled through technology and social media companies.”
The chairman requested Meta, which is the parent of Facebook, provide the committee with its internal policies and documents explaining how the Meta AI chatbot is designed, reviewed, managed and updated to incorporate newsworthy events and “to avoid query results that Meta desires to limit or preclude because it views them to be harmful or dangerous.”
The letter to Mr. Pichai cites reports from Google users that automated search prompts on assassination attempts produced results for failed attempts on former presidents, including Harry Truman, Gerald Ford and Ronald Reagan, and other historical figures like Archduke Franz Ferdinand, but omitted the far more recent attempt on Mr. Trump.
Google, whose parent is Alphabet, provided a staff-level briefing to the Oversight Committee last week, explaining that the autocomplete results omitted Mr. Trump’s assassination attempt “due to a safety protocol concerning predicted assassination attempts of current political leaders, and [that] Google had not yet updated the autocomplete feature to reflect that an assassination attempt of President Trump had occurred,” according to Mr. Comer’s letter.
The committee still has questions and is requesting Google turn over its internal policies and documents explaining how the company’s search function and autocomplete feature are designed, reviewed, managed and updated.
Mr. Comer requested Meta and Google turn over the requested documents by Aug. 28.
Sen. Roger Marshall sent a similar letter last month to Mr. Pichai about Google suppressing searches related to the assassination attempt on Mr. Trump.
The Kansas Republican on Wednesday released the response he received from Google, in which the company acknowledged that its autocomplete predictions “are not perfect” and did not initially provide prompts for queries about the attempt on Mr. Trump’s life.
“Our systems are designed to prohibit Autocomplete predictions for hypothetical political violence against current figures (as opposed to against historical figures such as President Truman, whom you asked about),” Mark Isakowitz, Google’s vice president for government affairs and public policy, said in the letter to Mr. Marshall.
Mr. Isakowitz said those system blockers were still in place in the immediate aftermath of the assassination attempt on Mr. Trump.
“We recognize that these out-of-date systems led to an inadequate user experience,” he said. “Once the issue was brought to our attention, we began working on improvements that have already started rolling out. To be clear, these were systematic improvements – not manual fixes – that affected a range of Autocomplete predictions.”
Mr. Marshall was dissatisfied with Google’s response, calling the company “a monopolistic agent of propaganda for the globalist Left.”
“Most shocking is their bizarre written defense that Trump’s assassination attempt was a ’hypothetical act of political violence’ even up to the point of our public inquiry on July 28th,” he said in a statement.
Mr. Marshall said he wants Google’s top executives to testify before the Senate Homeland Security and Governmental Affairs Committee.
“Under oath, they won’t be able to hide behind an algorithm or get away with corporate double-speak,” he said. “We are launching a full investigation into Google’s litany of offenses, time for accountability and consequence.”
• Lindsey McPherson can be reached at lmcpherson@washingtontimes.com.
Please read our comment policy before commenting.