Google Engineer Leaks Nearly 1,000 Pages of Internal Documents, Alleging Bias, Censorship

Google Engineer Leaks Nearly 1,000 Pages of Internal Documents, Alleging Bias, Censorship
Former Google software engineer Zach Vorhies in an undated file photo. (Courtesy of Project Veritas)
Petr Svab
8/14/2019
Updated:
8/20/2019

A former Google engineer has released nearly 1,000 pages of documents that he says prove that the company, at least in some of its products, secretly boosts or demotes content based on what it deems to be true or false, while publicly claiming to be a neutral platform.

The software engineer, Zach Vorhies, first provided the documents to Project Veritas, a right-leaning investigative journalism nonprofit, as well as the Justice Department’s antitrust division, which has been investigating Google for potentially anti-competitive behavior.

“I thought that our election system is going to be compromised forever by this company that told the American public that it was not going to do any evil,” he told Project Veritas in a video published Aug. 14. “And I saw that they were making really quick moves. ... They were intending to scope the information landscape so that they could create their own version of what was objectively true.”

Going Public

Vorhies said he worked for Google for eight years, making $260,000 a year, when counting in the gains from the Google stock he owns.

“I had every incentive in the world to stay at the company and just collect the paycheck,” he said, noting that most others would do that.

“But I could never live with myself knowing that, if Google was able to implement the plans that they were planning, that I, at the moment of choice, backed out because I was selfish.”

Vorhies first came to Project Veritas more than a month ago, disclosing some documents and answering questions with his face hidden and his voice disguised.

When he returned to work, however, Google sent him a letter demanding, among other things, that he turn over his employee badge and work laptop, which he did, and “cease and desist” from disclosing “any non-public Google files.” Afraid for his safety, he posted on Twitter that if something would happen to him, all the documents he took would be released to the public.

Google then did a “wellness check” on him, he said. The San Francisco police received a call that Vorhies may be mentally ill. A group of officers waited for him outside his house and put him in handcuffs. “This is a large way in which they intimidate their employees that go rogue on the company,” he said.

Vorhies then decided that it would be safer for him to go public.

Vorhies called Google a “political machine” bent on preventing anybody like President Donald Trump from getting elected again. He said there are other Google employees who “see what’s going on and they are really scared.”

Changes at the company that worried him started in 2016, he said.

The documents indicate that Google has ramped up emphasis on suppressing what it deems “fake news.” That has led it to review news content using a variety of manual and automated means to make calls on what is true and what is “misinformation” and sort results accordingly.

Most of the documents appear to pertain to Google News, an aggregator featured prominently at the top of the page for news-related search results.

Google News

One document describes “Project Purple Rain: Crisis Response & Escalation,” the goal of which is to establish “processes to detect and handle misinformation across products during crises” and “install 24/7 team of trained analysts ready to make policy calls and take actions across news surfaces including News, News 360 and Feed.”

“News” appears to pertain to “Google News” and “Feed,” a rebrand of the former “Google Now” product, showing news articles below the search bar on the Google mobile app.

Another document, a presentation that appears to date back to late 2017, explains that websites that apply to be included in Google News results need to pass an automated review that checks their technical parameters, and also a manual review of their “processes, policies, and editorial guidelines.” If accepted, the sites are then repeatedly checked and given “demotion penalties” for infractions.

Then, however, the presentation presents “potential” next steps, which included expanding its screening policies to cover “fringe/controversial” content, such as that which is “factually incorrect, fake, irrelevant.” Furthermore, the document suggests that Google should also address “sensitive” content, such as that which involves “hate,” “diversity,” and “bias” or is “geo-politically sensitive.”

One of the goals of the effort was a “clean & regularly sanitized news corpus,” it reads.

It’s not clear whether these steps have been implemented.

‘Fringe Ranking’

One of the documents says that Paul Haahr, Google’s principal engineer, leads the effort on “fringe ranking” with the goal of “not showing fake news, hate speech, conspiracy theories, or science/medical/history denial unless we’re sure that’s what the user wants.”
Combined with information from other documents, “fringe ranking” appears to mean that unless a user already knows what specifically to look for, Google will hide from the user anything it labels “fake news, hate speech, conspiracy theories, or science/medical/history denial.” Such information would then be effectively obscured from users who haven’t yet been exposed to it.

Feed Blacklist

Yet another document lists websites whose content is manually banned from showing up in the “Feed.”

It includes a number of fringe sites on the political right, the more prominent ones including thegatewaypundit.com, truepundit.com, and redstate.com.

It also includes some sites on the progressive left, including mediamatters.org, forwardprogressives.com, occupydemocrats.com, and learnprogress.org.

But it also includes relatively mainstream right-leaning sites such as dailycaller.com, louderwithcrowder.com, and newsbusters.org.

The list indicates it’s because those are “sites with high user block rate.” It’s not clear though why these sites would need to be blacklisted manually if a high block rate alone was to automatically place them on the list.

Videos Manually Rated

One presentation slide, apparently photographed from a computer screen, bears the title “Fake news & other fringe: Trashy recap” and says that “every day, top 250 videos [on YouTube] in top 26 locales are rated by multiple human raters” and that “Trashy filtering launched on [YouTube’s] Home [page], Search, Trending [list], and Suggested [videos list].” This has led to a 50 percent decline in user complaints, it states.
Tech companies that cave too easily to complaints have been mentioned by some conservatives as one of the underlying reasons for disproportionate censorship of right-leaning content. People on the political left are much more likely to call a variety of statements “hateful,” while those on the right tend to call the same statements “offensive, but not hateful,” a 2017 Cato survey found (pdf).
Google, as well as other major tech companies, prohibit content they consider “hate speech”—itself a subjective standard impossible to enforce fairly, according to Nadine Strossen, a law professor and former president of the American Civil Liberties Union.
Google hasn’t responded to a request for comment.

Bias

The documents Vorhies provided previously, together with his explanations and hidden camera recordings by Project Veritas of other Google employees, indicate that the company has created a concept of “fairness” through which it infuses the political preferences of its mostly left-leaning workforce into its products.
Several studies have shown that Google News, in particular, is biased to the left.

Google has repeatedly denied political bias in its products. Vorhies suggested, though, that Google tries to present itself as a neutral platform to preserve legal protection under Section 230, which shields internet services from liability for user-generated content.

“Google is playing both sides of the game,” he said. “On the one hand, they’re saying they are a platform and that they are immune from being sued for the content that they host on their website. On the other hand, they’re acting as a publisher, in which they’re determining the editorial agenda of these certain companies, and they are applying that. If people don’t fall in line with their editorial agenda, then their news articles get deboosted and deranked. And if people do fall in line with their editorial agenda, it gets boosted and pushed to the top.”

Robert Epstein, a psychologist who has spent years researching Google’s influence on its users, has published research showing that just by deciding the sequence of top search results, the company can sway undecided voters.
Epstein determined that this has led to 2.6 million votes shifting in the 2016 presidential election to Trump’s opponent, former Secretary of State Hillary Clinton. He warned that in 2020, if companies such as Google and Facebook all support the same candidate, they will be able to shift 15 million votes—well beyond the margin most presidents have won by.
Trump has reportedly been working on an executive order to address politically biased censorship by social media companies.
Correction: A previous version of this article inaccurately described the amount of documents that Zach Vorhies released. Vorhies released nearly 1,000 pages of documents, not nearly 1,000 documents. The Epoch Times regrets the error.