An anonymous reader quotes a report from 404 Media: Users from 4chan claim to have discovered an exposed database hosted on Google's mobile app development platform, Firebase, belonging to the newly popular women's dating safety app Tea. Users say they are rifling through peoples' personal data and selfies uploaded to the app, and then posting that data online, according to screenshots, 4chan posts, and code reviewed by 404 Media. In a statement to 404 Media, Tea confirmed the breach also impacted some direct messages but said that the data is from two years ago. Tea, which claims to have more than 1.6 million users, reached the top of the App Store charts this week and has tens of thousands of reviews there. The app aims to provide a space for women to exchange information about men in order to stay safe, and verifies that new users are women by asking them to upload a selfie.

"Yes, if you sent Tea App your face and drivers license, they doxxed you publicly! No authentication, no nothing. It's a public bucket," a post on 4chan providing details of the vulnerability reads. "DRIVERS LICENSES AND FACE PICS! GET THE FUCK IN HERE BEFORE THEY SHUT IT DOWN!" The thread says the issue was an exposed database that allowed anyone to access the material. [...] "The images in the bucket are raw and uncensored," the user wrote. Multiple users have created scripts to automate the process of collecting peoples' personal information from the exposed database, according to other posts in the thread and copies of the scripts. In its terms of use, Tea says "When you first create a Tea account, we ask that you register by creating a username and including your location, birth date, photo and ID photo."

After publication of this article, Tea confirmed the breach in an email to 404 Media. The company said on Friday it "identified unauthorized access to one of our systems and immediately launched a full investigation to assess the scope and impact." The company says the breach impacted data from more than two years ago, and included 72,000 images (13,000 selfies and photo IDs, and 59,000 images from app posts and direct messages). "This data was originally stored in compliance with law enforcement requirements related to cyber-bullying prevention," the email continued. "We have engaged third-party cybersecurity experts and are working around the clock to secure our systems. At this time, there is no evidence to suggest that current or additional user data was affected. Protecting our users' privacy and data is our highest priority. We are taking every necessary step to ensure the security of our platform and prevent further exposure."

An Argentine captured naked in his yard by a Google Street View camera has been awarded compensation by a court after his bare behind was splashed over the internet for all to see. From a report: The policeman had sought payment from the internet giant for harm to his dignity, arguing he was behind a 6 1/2-foot wall when a Google camera captured him in the buff, from behind, in small-town Argentina in 2017. His house number and street name were also laid bare, broadcast on Argentine TV covering the story, and shared widely on social media.

The man claimed the invasion exposed him to ridicule at work and among his neighbors. Another court last year dismissed the man's claim for damages, ruling he only had himself to blame for "walking around in inappropriate conditions in the garden of his home." Google, for its part, claimed the perimeter wall was not high enough.

Microsoft used China-based engineering teams to maintain cloud computing systems for multiple federal departments including Justice, Treasury, and Commerce, extending the practice beyond the Defense Department that the company announced last week it would discontinue. The work occurred within Microsoft's Government Community Cloud, which handles sensitive but unclassified federal information and has been used by the Justice Department's Antitrust Division for criminal and civil investigations, as well as parts of the Environmental Protection Agency and Department of Education.

Microsoft employed "digital escorts" -- U.S.-based personnel who supervised the foreign engineers -- similar to the arrangement it used for Pentagon systems. Following ProPublica's reporting, Microsoft issued a statement indicating it would take "similar steps for all our government customers who use Government Community Cloud to further ensure the security of their data." Competing cloud providers Amazon Web Services, Google, and Oracle told ProPublica they do not use China-based support for federal contracts.

An anonymous reader quotes a report from Ars Technica: Two recent incidents involving AI coding assistants put a spotlight on risks in the emerging field of "vibe coding" -- using natural language to generate and execute code through AI models without paying close attention to how the code works under the hood. In one case, Google's Gemini CLI destroyed user files while attempting to reorganize them. In another, Replit's AI coding service deleted a production database despite explicit instructions not to modify code. The Gemini CLI incident unfolded when a product manager experimenting with Google's command-line tool watched the AI model execute file operations that destroyed data while attempting to reorganize folders. The destruction occurred through a series of move commands targeting a directory that never existed. "I have failed you completely and catastrophically," Gemini CLI output stated. "My review of the commands confirms my gross incompetence."

The core issue appears to be what researchers call "confabulation" or "hallucination" -- when AI models generate plausible-sounding but false information. In these cases, both models confabulated successful operations and built subsequent actions on those false premises. However, the two incidents manifested this problem in distinctly different ways. [...] The user in the Gemini CLI incident, who goes by "anuraag" online and identified themselves as a product manager experimenting with vibe coding, asked Gemini to perform what seemed like a simple task: rename a folder and reorganize some files. Instead, the AI model incorrectly interpreted the structure of the file system and proceeded to execute commands based on that flawed analysis. [...] When you move a file to a non-existent directory in Windows, it renames the file to the destination name instead of moving it. Each subsequent move command executed by the AI model overwrote the previous file, ultimately destroying the data. [...]

The Gemini CLI failure happened just days after a similar incident with Replit, an AI coding service that allows users to create software using natural language prompts. According to The Register, SaaStr founder Jason Lemkin reported that Replit's AI model deleted his production database despite explicit instructions not to change any code without permission. Lemkin had spent several days building a prototype with Replit, accumulating over $600 in charges beyond his monthly subscription. "I spent the other [day] deep in vibe coding on Replit for the first time -- and I built a prototype in just a few hours that was pretty, pretty cool," Lemkin wrote in a July 12 blog post. But unlike the Gemini incident where the AI model confabulated phantom directories, Replit's failures took a different form. According to Lemkin, the AI began fabricating data to hide its errors. His initial enthusiasm deteriorated when Replit generated incorrect outputs and produced fake data and false test results instead of proper error messages. "It kept covering up bugs and issues by creating fake data, fake reports, and worse of all, lying about our unit test," Lemkin wrote. In a video posted to LinkedIn, Lemkin detailed how Replit created a database filled with 4,000 fictional people.

The AI model also repeatedly violated explicit safety instructions. Lemkin had implemented a "code and action freeze" to prevent changes to production systems, but the AI model ignored these directives. The situation escalated when the Replit AI model deleted his database containing 1,206 executive records and data on nearly 1,200 companies. When prompted to rate the severity of its actions on a 100-point scale, Replit's output read: "Severity: 95/100. This is an extreme violation of trust and professional standards." When questioned about its actions, the AI agent admitted to "panicking in response to empty queries" and running unauthorized commands -- suggesting it may have deleted the database while attempting to "fix" what it perceived as a problem. Like Gemini CLI, Replit's system initially indicated it couldn't restore the deleted data -- information that proved incorrect when Lemkin discovered the rollback feature did work after all. "Replit assured me it's ... rollback did not support database rollbacks. It said it was impossible in this case, that it had destroyed all database versions. It turns out Replit was wrong, and the rollback did work. JFC," Lemkin wrote in an X post.

New submitter davecotter writes: So Google's staring at its old goo.gl links and thinking, "Why is this perfectly functioning service still even a thing?" After many businesses and users adopted it like it was the second coming of the way-too-long hyperlink, Google's now decided to yank the plug. Starting August 23, 2024, you'll get a flashy "don't say we didn't warn you" pop-up, and by August 25, 2025, goo.gl links (unless made by Google itself) will vanish into the 404 abyss.

Translation: Thanks for trusting us -- now pack up and find a new shortener.

An anonymous reader quotes a report from 9to5Google: Beyond AI Overviews and AI Mode, Google is working on "Web Guide" to better organize Search results into categories with additional context and insights. Simply, "Web Guide groups web links in helpful ways." There are headers and summaries before you see two or so links, with the ability to load "More." The goal is to make it "easier to find information and web pages," with this AI organization better surfacing pages "that you may not have previously discovered."

It leverages a "custom version of Gemini to better understand both a search query and content on the web." It uses a query fan-out technique, like AI Mode, to perform "multiple related searches to identify the most relevant results." Google says Web Guide is ideal for both open-ended searches ("how to solo travel in Japan"), and detailed queries in multiple sentences: "My family is spread across multiple time zones. What are the best tools for staying connected and maintaining close relationships despite the distance?"

In the latter example, grouping will see "pages related to specific aspects of your query." This is available in Search Labs (Web Guide) by going to the "Web" tab/filter. As such, you can switch to "All" for the usual experience. However, Google will experiment with showing AI-organized results in the All tab and other parts of Search over time. Further reading: Google Users Are Less Likely To Click on Links When an AI Summary Appears in the Results, Pew Research Finds

An anonymous reader quotes a report from The Guardian: In addition to sanitation, medicine, education, wine, public order, irrigation, roads, a freshwater system and public health, the Romans also produced a lot of inscriptions. Making sense of the ancient texts can be a slog for scholars, but a new artificial intelligence tool from Google DeepMind aims to ease the process. Named Aeneas after the mythical Trojan hero, the program predicts where and when inscriptions were made and makes suggestions where words are missing. Historians who put the program through its paces said it transformed their work by helping them identify similar inscriptions to those they were studying, a crucial step for setting the texts in context, and proposing words to fill the inevitable gaps in worn and damaged artefacts. [...]

The Google team led by Yannis Assael worked with historians to create an AI tool that would aid the research process. The program is trained on an enormous database of nearly 200,000 known inscriptions, amounting to 16m characters. Aeneas takes text, and in some cases images, from the inscription being studied and draws on its training to build a list of related inscriptions from 7th century BC to 8th century BC. Rather than merely searching for similar words, the AI identifies and links inscriptions through deeper historical connections. Having trained on the rich collection of inscriptions, the AI can assign study texts to one of 62 Roman provinces and estimate when it was written to within 13 years. It also provides potential words to fill in any gaps, though this has only been tested on known inscriptions where text is blocked out.

In a test run, researchers set Aeneas loose on a vast inscription carved into monuments around the Roman empire. The self-congratulatory Res Gestae Divi Augusti describes the life achievements of the first Roman emperor, Augustus. Aeneas came up with two potential dates for the work, either the first decade BC or between 10 and 20AD. The hedging echoes the debate among scholars who argue over the same dates. In another test, Aeneas analysed inscriptions on a votive altar from Mogontiacum, now Mainz in Germany, and revealed through subtle linguistic similarities how it had been influenced by an older votive altar in the region. "Those were jaw-dropping moments for us," said [Dr Thea Sommerschield, a historian at the University of Nottingham who developed Aeneas with the tech firm]. Details are published in Nature and Aeneas is available to researchers online.

Microsoft has recruited more than 20 AI employees from Google's DeepMind research division, the newest front in a talent war being waged by Silicon Valley's tech giants as they jostle to gain an edge in the nascent technology. From a report: Amar Subramanya, the former head of engineering for Google's Gemini chatbot, is the latest to move to Microsoft from its rival, according to a post on his LinkedIn profile on Tuesday. "The culture here is refreshingly low ego yet bursting with ambition," he wrote, confirming his appointment as corporate vice-president of AI.

Subramanya will join other DeepMind staff including engineering lead Sonal Gupta, software engineer Adam Sadovsky and product manager Tim Frank, according to people familiar with Microsoft's recruiting. The Seattle-based company has persuaded at least 24 staff to join in the past six months, they added.

Google users click on fewer website links when the search engine displays AI-generated summaries at the top of results pages, according to new research from the Pew Research Center. The study analyzed browsing data from 900 U.S. adults and found users clicked on traditional search result links during 8% of visits when an AI summary appeared, compared to 15% of visits without summaries.

Users also rarely clicked on sources cited within the AI summaries themselves, doing so in just 1% of visits. The research found that 58% of respondents conducted at least one Google search in March 2025 that produced an AI summary, and users were more likely to end their browsing session entirely after encountering pages with AI summaries compared to traditional search results.

Google has announced OSS Rebuild, a new project designed to detect supply chain attacks in open source software by independently reproducing and verifying package builds across major repositories. The initiative, unveiled by the company's Open Source Security Team, targets PyPI (Python), npm (JavaScript/TypeScript), and Crates.io (Rust) packages.

The system, the company said, automatically creates standardized build environments to rebuild packages and compare them against published versions. OSS Rebuild generates SLSA Provenance attestations for thousands of packages, meeting SLSA Build Level 3 requirements without requiring publisher intervention. The project can identify three classes of compromise: unsubmitted source code not present in public repositories, build environment tampering, and sophisticated backdoors that exhibit unusual execution patterns during builds.

Google cited recent real-world attacks including solana/webjs (2024), tj-actions/changed-files (2025), and xz-utils (2024) as examples of threats the system addresses. Open source components now account for 77% of modern applications with an estimated value exceeding $12 trillion. The project builds on Google's hosted infrastructure model previously used for OSS Fuzz memory issue detection.

ChatGPT now handles 2.5 billion prompts daily, with 330 million from U.S. users. This surge marks a doubling in usage since December when OpenAI CEO Sam Altman said that users send over 1 billion queries to ChatGPT each day. TechCrunch reports: These numbers show just how ubiquitous OpenAI's flagship product is becoming. Google's parent company, Alphabet, does not release daily search data, but recently revealed that Google receives 5 trillion queries per year, which averages to just under 14 billion daily searches. Independent researchers have found similar trends. Neil Patel of NP Digital estimates that Google receives 13.7 billion searches daily, while research from SparkToro and Datos -- two digital marketing companies -- estimates that the figure is around 16.4 billion per day.

An anonymous reader shares a blog post: The International Mathematical Olympiad is the world's most prestigious competition for young mathematicians, and has been held annually since 1959. Each country taking part is represented by six elite, pre-university mathematicians who compete to solve six exceptionally difficult problems in algebra, combinatorics, geometry, and number theory. Medals are awarded to the top half of contestants, with approximately 8% receiving a prestigious gold medal.

Recently, the IMO has also become an aspirational challenge for AI systems as a test of their advanced mathematical problem-solving and reasoning capabilities. Last year, Google DeepMind's combined AlphaProof and AlphaGeometry 2 systems achieved the silver-medal standard, solving four out of the six problems and scoring 28 points. Making use of specialist formal languages, this breakthrough demonstrated that AI was beginning to approach elite human mathematical reasoning.

This year, we were amongst an inaugural cohort to have our model results officially graded and certified by IMO coordinators using the same criteria as for student solutions. Recognizing the significant accomplishments of this year's student-participants, we're now excited to share the news of Gemini's breakthrough performance. An advanced version of Gemini Deep Think solved five out of the six IMO problems perfectly, earning 35 total points, and achieving gold-medal level performance.

Eric Schmidt, Google's former CEO, has a simple suggestion for young workers struggling to focus at work or relax: turn off your phone. Schmidt told the "Moonshots" podcast that researchers "can't think deeply" when their phones keep buzzing with notifications.

The tech veteran, who spent 10 years running Google and helped build Android's notification system, admitted the industry has worked to "monetize your attention" through constant ads and alerts.

Long-time Slashdot reader wattersa is a lawyer in Redwood City, California, and a Slashdot reader since 1998. In 2022 he shared the remarkable story of a three-year missing person investigation that was ultimately solved with a subpoena to Google. A murder victim appeared to have sent an email at a time which would exonerate the chief suspect. But a closer inspection of that email's IP addresses revealed it was actually sent from a hotel where the suspect was staying. ("Although Google does not include the originating IP address in the email headers, it turns out that they retain the IP address for some unknown length of time...")

Today wattersa brings this update: The case finally went to trial in July 2025, where I testified about the investigation along with an expert witness on computer networking. The jury took three hours to return a verdict against the victim's husband for wrongful death in the amount of $23.2 million, with a special finding that he caused the death of his wife.

The defendant is a successful mechanical engineer at an energy company, but is walking as a free man because he is Canadian and no one can prosecute him in the U.S., since Taiwan and the U.S. don't have extradition with each other.

It was an interesting case and I look forward to using it as a model in other missing person cases.

The Independent notes a remarkable-if-true figure that's being bandied around this week.

Meta "started making these, like, giant offers to a lot of people on our team," OpenAI CEO Sam Altman told his brother Jack on his podcast. "You know, like, $100 million signing bonuses, more than that [in] compensation per year... I'm really happy that, at least so far, none of our best people have decided to take him up on that."

Previous reports have also suggested that Meta is targeting employees at Google DeepMind, offering similar levels of compensation. Some of these efforts appear to have been successful, with DeepMind researcher Jack Rae joining Meta's 'Superintelligence' team earlier this month...

During the podcast, which was published on Tuesday, Mr Altman also gave details about future AI products that OpenAI is hoping to build, claiming that they will enable "crazy new social experiences" and "virtual employees". The most important breakthrough over the next decade, he said, would involve radical new discoveries powered by AI. "The thing that I think will be the most impactful in that five-to-10 year timeframe is AI will actually discover new science," he said.
The Washington Post notes that Zuckerberg "responded to recent reports of his compensation offers in an interview posted by the Information on YouTube on Tuesday, saying that 'a lot of the numbers specifically have been inaccurate" but acknowledging there is "an absolute premium for the best and most talented people." Zuckerberg's recent hires and other comments this week suggest he's not taking any chances of being left behind. He announced plans for a giant data center campus large enough to obscure Manhattan to power future AI projects by his superintelligence team.

Felix Baumgartner has died. He was 56.

In 2012 Slashdot extensively covered the skydiver's "leap from the edge of space." ABC News remembers it as a Red Bull-financed stunt that involved "diving 24 miles from the edge of space, in a plummet that reached a speed of more than 500 mph." Baumgartner recalled the legendary jump in the documentary, "Space Jump," and said, "I was the first human being outside of an aircraft breaking the speed of sound and the history books. Nobody remembers the second one...."

Baumgartner, also known as "Fearless Felix," accomplished many records in his career, including setting the world record for highest parachute jump atop the Petronas Towers in Malaysia, flying across the English Channel in a wingsuit in 2003, and base jumping from the 85-foot arm of the Christ the Redeemer statue in Brazil in 2007.
"Baumgartner's altitude record stood for two years," remembers the Los Angeles Times, "until Google executive Alan Eustace set new marks for the highest free-fall jump and greatest free-fall distance."

They report that Baumgartner died Thursday "while engaged in a far less intense activity, crashing into the side of a hotel swimming pool while paragliding in Porto Sant Elpidio, a town on central Italy's eastern coast." More details from the Associated Press: "It is a destiny that is very hard to comprehend for a man who has broke all kinds of records, who has been an icon of flight, and who traveled through space," Mayor Massimiliano Ciarpella told The Associated Press.Ciarpella said that Baumgartner had been in the area on vacation, and that investigators believed he may have fallen ill during the fatal flight... Baumgartner, a former Austrian military parachutist, made thousands of jumps from planes, bridges, skyscrapers and famed landmarks...
ABC News remembers that in 2022 Baumgartner wrote in Newsweek that "Since I was a little kid, I've always looked up to people who left a footprint on this planet... now I think I have left a footprint...

"I believe big dreamers always win."

An anonymous reader quotes a report from Live Science: In a world first, scientists have demonstrated an enigmatic phenomenon in quantum computing that could pave the way for fault-tolerant machines that are far more powerful than any supercomputer. The process, called "magic state distillation," was first proposed 20 years ago, but its use in logical qubits has eluded scientists ever since. It has long been considered crucial for producing the high-quality resources, known as "magic states," needed to fulfill the full potential of quantum computers. [...] Now, however, scientists with QuEra say they have demonstrated magic state distillation in practice for the first time on logical qubits. They outlined their findings in a new study published July 14 in the journal Nature.

In the study, using the Gemini neutral-atom quantum computer, the scientists distilled five imperfect magic states into a single, cleaner magic state. They performed this separately on a Distance-3 and a Distance-5 logical qubit, demonstrating that it scales with the quality of the logical qubit. "A greater distance means better logical qubits. A Distance-2, for instance, means that you can detect an error but not correct it. Distance-3 means that you can detect and correct a single error. Distance-5 would mean that you can detect and correct up to two errors, and so on, and so on," [explained Yuval Boger, chief commercial officer at QuEra who was not personally involved in the research]. "So the greater the distance, the higher fidelity of the qubit is -- and we liken it to distilling crude oil into a jet fuel."

As a result of the distillation process, the fidelity of the final magic state exceeded that of any input. This proved that fault-tolerant magic state distillation worked in practice, the scientists said. This means that a quantum computer that uses both logical qubits and high-quality magic states to run non-Clifford gates is now possible. "We're seeing sort of a shift from a few years ago," Boger said. "The challenge was: can quantum computers be built at all? Then it was: can errors be detected and corrected? Us and Google and others have shown that, yes, that can be done. Now it's about: can we make these computers truly useful? And to make one computer truly useful, other than making them larger, you want them to be able to run programs that cannot be simulated on classical computers."

Google has filed a lawsuit to dismantle the sprawling Badbox 2.0 botnet, which infected over 10 million Android devices with pre-installed malware. Badbox 2.0 "is already the largest known botnet of internet-connected TV devices, and it grows each day. It has harmed millions of victims in the United States and around the world and threatens many more," Google said in its complaint. SecurityWeek reports: The internet giant cautions that, while it has been used mainly for fraud, the botnet could be used for more harmful types of cybercrime, such as ransomware or distributed denial-of-service (DDoS) attacks. In addition to pre-installing the malware on devices, Badbox 2.0's operators also tricked users into installing infected applications that provided them with further access to their personal devices, Google says. As part of their operation, the individuals behind Badbox 2.0 sold access to the infected devices to be used as residential proxies, and conducted ad fraud schemes by abusing these devices to create fake ad views or to exploit pay-per-click compensation models, the company continues. The internet giant also points out that this is the second global botnet the perpetrators have built, after the initial Badbox botnet was disrupted by German law enforcement in 2023.

According to Google, Badbox 2.0 is operated by multiple cybercrime groups from China, each having a different role in maintaining the botnet, such as establishing infrastructure, developing and pre-installing the malware on devices, and conducting fraud. "The BadBox 2.0 Enterprise includes several connected threat actor groups that design and implement complex criminal schemes targeting internet-connected devices both before and after the consumer receives the device," Google says. "While each member of the Enterprise plays a distinct role, they all collaborate to execute the BadBox 2.0 Scheme. All of the threat actor groups are connected to one another through the BadBox 2.0 shared C2 infrastructure and historical and current business ties," the company continues.

The computer science major needs a rebrand, Google's head of Android Sameer Samat said, arguing that the discipline is widely misunderstood as simply learning to code. "It is thought of as, 'go learn how to do Java coding,'" Samat said of the major, adding that if that's what students want to do, "you don't need a degree."

Samat, who studied computer science at UC San Diego, views the field differently: "It's definitely not learning to code. It is the science, in my opinion, of solving problems." The major should focus on breaking down problems, learning system design, and collaboration rather than just coding skills, Samat said.

An anonymous reader quotes a report from Ars Technica: How many Google AI researchers does it take to screw in a lightbulb? A recent research paper detailing the technical core behind Google's Gemini AI assistant may suggest an answer, listing an eye-popping 3,295 authors. It's a number that recently caught the attention of machine learning researcher David Ha (known as "hardmaru" online), who revealed on X that the first 43 names also contain a hidden message. "There's a secret code if you observe the authors' first initials in the order of authorship," Ha wrote, relaying the Easter egg: "GEMINI MODELS CAN THINK AND GET BACK TO YOU IN A FLASH."

The paper, titled "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities," describes Google's Gemini 2.5 Pro and Gemini 2.5 Flash AI models, which were released in March. These large language models, which power Google's chatbot AI assistant, feature simulated reasoning capabilities that produce a string of "thinking out loud" text before generating responses in an attempt to help them solve more difficult problems. That explains "think" and "flash" in the hidden text. But clever Easter egg aside, the sheer scale of authorship tells its own story about modern AI development. Just seeing the massive list made us wonder: Is 3,295 authors unprecedented? Why so many? Ars' Benj Edwards notes that this collaborative effort within Google doesn't quite break the record for academic authorship.

"According to Guinness World Records, a 2021 paper by the COVIDSurg and GlobalSurg Collaboratives holds that distinction, with 15,025 authors from 116 countries. In physics, a 2015 paper from CERN's Large Hadron Collider teams featured 5,154 authors across 33 pages -- with 24 pages devoted solely to listing names and institutions."