Google terminates Dr. Timnit Gebru after research censorship
Dr. Timnit Gebru, a Black woman research scientist and co-lead of Google's Ethical AI team, was terminated on December 2, 2020, after Google censored her research on large language models. Google claimed she resigned, but she alleges it was retaliation for her pushback against censorship and workplace discrimination. The termination sparked widespread support from Googlers and academics.
- Company involved
- Google
10 source articles · read the reporting →
Teenager dies after Google Maps directs him onto abandoned Siberian road
An 18-year-old driver, Sergey Ustinov, froze to death in Siberia after Google Maps directed him onto a disused shortcut on the Road of Bones. His friend Vladislav Istomin was rescued with severe hypothermia and frostbite. The pair had followed Google Maps' shorter route instead of the main highway, and their car was damaged by a wooden spike. The incident is under investigation by Russian authorities.
- Company involved
- Google
- AI system involved
- Google Maps
7 source articles · read the reporting →
Google's secret blocklist fails to block many hate terms on YouTube ads
The Markup investigated Google's secret blocklist that hides YouTube hate videos from advertisers. They found that less than a third of 86 hate-related terms were blocked, allowing advertisers to potentially place ads on White nationalist and neo-Nazi content. Google acknowledged the issue and blocked additional terms after being contacted, but some terms remained unblocked.
- Company involved
- Google
- AI system involved
- Google Ads blocklist
10 source articles · read the reporting →
Google, Amazon, Microsoft image AI shows gender bias, labelling women by appearance and men by profession
A study by US and European researchers found that Google, Amazon, and Microsoft's image recognition services applied three times as many appearance-related labels to photos of women as to men. The top labels for men were 'official' and 'businessperson', while for women they were 'smile' and 'chin'. Google acknowledged the issue and had previously switched off gender detection, but tests by WIRED suggested the bias persisted. The researchers warn that such biased AI could obscure women professionals and create a false image of reality.
- Company involved
- Google, Amazon, Microsoft
- AI system involved
- Google Cloud Vision, Amazon Rekognition, Microsoft image recognition service
1 source article · read the reporting →
Google image search for 'Jewish baby strollers' yields anti-Semitic oven images
In September 2020, a Google image search for 'Jewish baby strollers' returned rows of portable ovens, an offensive Holocaust allusion. Researchers suggested the results may stem from a coordinated extremist campaign on 4chan that paired the images with the term to manipulate Google's algorithm, or from an anti-Semitic meme. Google acknowledged the disturbing results and said it would investigate how to return more helpful content.
- Company involved
- Google
- AI system involved
- Google Image Search
2 source articles · read the reporting →
Google's ad portal suggested porn keywords for 'Black girls'
Google's Keyword Planner, used by advertisers to select search terms, suggested hundreds of pornographic keywords when users searched for 'Black girls', 'Latina girls', and 'Asian girls', but returned no suggestions for 'White girls' or 'White boys'. The Markup reported that Google's own adult filter flagged many of these terms, but the system still recommended them. Google acknowledged the issue, removed the terms, and said it was working to prevent recurrence.
- Company involved
- Google
- AI system involved
- Google Ads Keyword Planner
9 source articles · read the reporting →
Hawley Slams Gmail’s Bias Against Republican Candidates, Demands Answers From Google - U.S. Senate Website (.gov)
Gmail's spam filter marked Republican fundraising emails as spam, preventing them from reaching intended recipients.
- Company involved
- Google
- AI system involved
- Gmail spam filtering algorithm
1 source article · read the reporting →
YouTube sued by LGBT creators over alleged algorithmic discrimination
A class action lawsuit filed by 12 LGBT YouTubers alleges that YouTube's algorithm discriminates against their content by demonetizing, age-restricting, and limiting the visibility of videos. The plaintiffs claim the automated system makes decisions within moments of upload, without human review, and that YouTube has failed to address the bias. The lawsuit, which includes a recording of a YouTube manager citing 'the gay thing' as a reason for restricting content, seeks systemic change and transparency.
- Company involved
- YouTube
1 source article · read the reporting →
Google to Stop Using Assistant Recordings Without Consent After Privacy Scandal
Google announced it will no longer use Assistant voice recordings to improve its AI without explicit user permission, following a scandal where contractors reviewed private conversations. The company paused human review of recordings in July 2019 and will now require users to opt in, with automatic deletion of old recordings. The move aims to restore trust after the privacy violation.
- Company involved
- Google
- AI system involved
- Google Assistant
1 source article · read the reporting →
Google contractors targeted homeless people for Pixel 4 facial recognition data
Google hired Randstad to collect facial data to train the Pixel 4's face recognition system. Contractors allegedly targeted homeless people and unaware students, using deceptive tactics such as calling it a "selfie game" and offering $5 without informing subjects they were being recorded. Google suspended the research program and opened an investigation following the report.
- Company involved
- Google
- AI system involved
- Pixel 4 facial recognition
1 source article · read the reporting →
YouTube algorithm radicalized Christchurch shooter leading to mosque attack
In March 2019, a terrorist killed 51 worshippers in two New Zealand mosques. A New Zealand government report found that YouTube's recommendation algorithm was a significant source of radicalization for the shooter, who regularly watched extremist content. YouTube has since taken steps to remove hate speech content but researchers say the underlying business model and opaque algorithms still enable radicalization.
- Company involved
- YouTube (Google)
- AI system involved
- YouTube recommendation algorithm
10 source articles · read the reporting →
Google's algorithm blurs protest graffiti on Hong Kong Street View
Google Street View's automatic blurring technology obscured two protest graffiti messages on Nathan Road in Hong Kong, including one calling for the death of Xi Jinping and another calling for liberation. Google said the blurring was an unintended algorithmic error, not deliberate censorship, and the messages remained partly visible from certain angles. The company acknowledged the flaw but did not indicate whether it would correct the images.
- Company involved
- Google
- AI system involved
- Google Street View
7 source articles · read the reporting →
Gmail Smart Reply prototype suggested 'Sent from my iPhone'
During testing of Gmail's Smart Reply feature, Google engineers discovered the algorithm had identified the phrase 'Sent from my iPhone' as a popular response to emails. The issue was fixed before the phrase became a suggested reply to users. The article also notes the prototype had a propensity to suggest 'I love you' in many contexts, which was also corrected.
- Company involved
- Google
- AI system involved
- Gmail Smart Reply
10 source articles · read the reporting →
Google Will Not Renew Project Maven Military Drone AI Contract
Google announced it will not renew its contract for Project Maven, a Pentagon program using AI to analyse drone imagery. The decision followed internal protests, with nearly a dozen employees resigning and thousands signing a letter declaring Google should not be in the business of war. The company will continue work until March 2019 but has not ruled out future military AI contracts.
- Company involved
- U.S. Department of Defense
- AI system involved
- Project Maven
1 source article · read the reporting →
Google Duplex AI assistant to identify itself as robot after criticism
Google demonstrated its Duplex AI assistant making lifelike phone calls without disclosing it was a robot, sparking accusations of deceit. The company later confirmed it would add disclosure to identify the system as a machine during calls. The feature was not yet a finished product at the time.
- Company involved
- Google
- AI system involved
- Google Duplex
10 source articles · read the reporting →
Waze Directed Driver into Lake Champlain in Vermont
A driver in Vermont alleges that the Waze navigation app directed him to turn onto a boat launch near a Coast Guard station, causing his car to slide about 100 feet onto Lake Champlain. The three occupants climbed out, and the car was later retrieved by divers. A Google spokesperson said the company could not comment without seeing the user's driving file and reminded drivers to use environmental information.
- Company involved
- Waze
- AI system involved
- Waze
3 source articles · read the reporting →
YouTube autocomplete algorithm suggested child sexual abuse phrases
YouTube's search autofill algorithm suggested pedophiliac phrases such as 'how to have s*x with your kids' to users searching for 'how to have'. Dozens of users reported the disturbing results. YouTube acknowledged the issue, removed the autocomplete suggestions, and said it is investigating the cause, which may be a coordinated trolling campaign.
- Company involved
- YouTube
- AI system involved
- YouTube autocomplete algorithm
10 source articles · read the reporting →
Google's Cloud Natural Language API Exhibits Bias Against Minority Groups
In October 2017, Google's Cloud Natural Language API, a sentiment analysis tool, was found to assign negative scores to phrases like 'I'm a homosexual' and 'I'm black', while giving positive scores to 'I'm straight' and neutral to 'White power'. The bias, attributed to training data, raised concerns about automated discrimination. Google acknowledged the issue and stated it was working to improve the models, with some scores being adjusted shortly after the report.
- Company involved
- Google
- AI system involved
- Cloud Natural Language API
6 source articles · read the reporting →
Google Home Mini defect caused continuous recording of user's conversations
Author Artem Russakovskii discovered that his Google Home Mini was recording all sounds in his bathroom 24/7 and sending them to Google servers due to a hardware defect causing phantom touch events. The device activated thousands of times per day without the hotword. Google acknowledged the issue, issued a software update to disable the top touch functionality, and removed all affected recordings from their servers. The company stated the defect affected a small number of early release devices.
- Company involved
- Google
- AI system involved
- Google Home Mini
6 source articles · read the reporting →
Google allowed ads targeting racist search terms, disabled them after inquiry
BuzzFeed News discovered that Google's advertising platform allowed a campaign to target people searching racist phrases, and that Google's keyword tool suggested additional offensive terms. The campaign served 17 ad impressions before Google disabled most of the keywords after being contacted. Google acknowledged the breach of its policies and said it had turned off the suggestions.
- Company involved
- Google
10 source articles · read the reporting →
Google's Perspective API Found to Exhibit Racial Bias in Toxicity Scores
In August 2017, Robyn Speer demonstrated that Google's Perspective API, a toxicity classifier, assigned higher toxicity scores to names associated with Black people and to terms like 'Mexican food' and 'Islam'. The system, intended to fight online toxicity, was found to replicate biases from its training data. Google's team later acknowledged the issue but attributed it to out-of-vocabulary word handling, and made superficial changes to the demo without addressing the underlying model bias.
- Company involved
- Google
- AI system involved
- Perspective API
8 source articles · read the reporting →
University of Chicago Medical Center sued over sharing patient data with Google
The University of Chicago Medical Center is accused in a class-action lawsuit of sharing hundreds of thousands of patient medical records with Google without proper de-identification or patient consent. The data, from patients treated between 2009 and 2016, allegedly included dates and provider notes that could allow re-identification. The lawsuit claims Google sought the records to develop its own electronic health record system and predictive models. The hospital and Google deny wrongdoing, stating the research partnership was legal and compliant with HIPAA.
- Company involved
- University of Chicago Medical Center
1 source article · read the reporting →
Google search algorithm returned Holocaust denial as top result
Google's search algorithm returned Holocaust denial content as the top result for the query 'did the Holocaust happen'. The result linked to Stormfront, a neo-Nazi site. Google later modified some results without public acknowledgment. The article alleges that Google's algorithm is being gamed by extremist sites.
- Company involved
- Google
- AI system involved
- Google Search
10 source articles · read the reporting →
Google used 11,000 novels without permission to train AI
Google researchers fed 11,000 novels, including works by authors Rebecca Forster and Erin McCarthy, into a neural network to improve conversational AI without seeking permission. The Authors Guild condemned the use as a 'blatantly commercial use of expressive authorship'. Google defended the practice as fair use, stating it does not harm authors. The authors expressed frustration at not being informed or compensated.
- Company involved
- Google
10 source articles · read the reporting →
