sciencehabit shares a report from Science Magazine: A hacked message in a streamed song makes Alexa send money to a foreign entity. A self-driving car crashes after a prankster strategically places stickers on a stop sign so the car misinterprets it as a speed limit sign. Fortunately these haven't happened yet, but hacks like this, sometimes called adversarial attacks, could become commonplace -- unless artificial intelligence (AI) finds a way to outsmart them. Now, researchers have found a new way to give AI a defensive edge. The work could not only protect the public. It also helps reveal why AI, notoriously difficult to understand, falls victim to such attacks in the first place. Because some AIs are too smart for their own good, spotting patterns in images that humans can't, they are vulnerable to those patterns and need to be trained with that in mind, the research suggests.

To identify this vulnerability, researchers created a special set of training data: images that look to us like one thing, but look to AI like another -- a picture of a dog, for example, that, on close examination by a computer, has catlike fur. Then the team mislabeled the pictures -- calling the dog picture an image of a cat, for example -- and trained an algorithm to learn the labels. Once the AI had learned to see dogs with subtle cat features as cats, they tested it by asking it to recognize fresh, unmodified images. Even though the AI had been trained in this odd way, it could correctly identify actual dogs, cats, and so on nearly half the time. In essence, it had learned to match the subtle features with labels, whatever the obvious features. The training experiment suggests AIs use two types of features: obvious, macro ones like ears and tails that people recognize, and micro ones that we can only guess at. It further suggests adversarial attacks aren't just confusing an AI with meaningless tweaks to an image. In those tweaks, the AI is smartly seeing traces of something else. An AI might see a stop sign as a speed limit sign, for example, because something about the stickers actually makes it subtly resemble a speed limit sign in a way that humans are too oblivious to comprehend. Engineers could change the way they train AI to help outsmart adversarial attacks. When the researchers trained an algorithm on images without the subtle features, "their image recognition software was fooled by adversarial attacks only 50% of the time," reports Science Magazine. "That compares with a 95% rate of vulnerability when the AI was trained on images with both obvious and subtle patterns."

Lenovo is launching a new AR-and-VR system targeted at businesses. It's called ThinkReality, and from the looks and description of the device and platform, it looks like a competitor for Microsoft's HoloLens. Engadget reports: There are two parts to the new ThinkReality system -- the AR headset and a software platform. The ThinkReality A6 is a comfortable 380-gram headset with two fisheye cameras on the front, as well as depth sensors and a 13-megapixel RGB sensor. There's also microphones onboard for voice control, and the headset can also detect where you're gazing to optimize resolution or navigation. You'll also be able to interact with your virtual environments using an included 3DoF controller. Untethered, the A6 can last up to four hours with its 6,800mAh battery, and you can still use the device while it's plugged in and charging.

The headset connects to an SSD-sized compute box that contains a Snapdragon 845 CPU running an Android-based platform. There's also an Intel Movidius chips powering waveguide optics here, and each eye on the A6 offers a 40-degree diagonal field of view and 1080p resolution. By comparison, the HoloLens 2 uses a Snapdragon 850 CPU and packs two 2K MEMS displays. Microsoft also squeezes all the computing components into the headset rather than in a separate box like Lenovo does. While I'm not a fan of having to carry around an additional accessory to power a headset, this setup does make the ThinkReality A6 lighter, so it's a compromise I'm willing to make. You can wear the box on a belt clip or an armband, which should make it easy to move around when wearing this setup at work. Lenovo claims this is "one of the lightest fully featured AR headsets in its class," and during a brief trial with a non-working model, I certainly found the A6 lightweight.

An anonymous reader quotes a report from Ars Technica: A U.S. Navy memo warns that 5G mobile networks are likely to interfere with weather satellites, and senators are urging the Federal Communications Commission to avoid issuing new spectrum licenses to wireless carriers until changes are made to prevent harms to weather forecasting. The FCC has already begun an auction of 24GHz spectrum that would be used in 5G networks. But Sens. Ron Wyden (D-Ore.) and Maria Cantwell (D-Wash.) today wrote a letter to FCC Chairman Ajit Pai, asking him to avoid issuing licenses to winning bidders "until the FCC approves the passive band protection limits that the National Aeronautics and Space Administration (NASA) and the National Oceanic and Atmospheric Administration (NOAA) determine are necessary to protect critical satellite-based measurements of atmospheric water vapor needed to forecast the weather."

The internal Navy memo on the topic, written on March 27 by U.S. Naval Observatory Superintendent Marc Eckardt, was made public by Wyden and Cantwell today. The Navy memo cited NOAA and NASA studies on interference from 24GHz spectrum, which is intended for mobile use and is adjacent to spectrum used for weather operations. "[A]s such, it is expected that interference will result in a partial-to-complete loss of remotely sensed water-vapor measurements," the Navy memo said. "It is also expected that impacts will be concentrated in urban areas of the United States first." The problem could affect Navy and Marine Corps forecasts of tropical cyclones as well as rain, ice, and snow, the memo said. The Navy memo recommends asking the FCC to "tighten out-of-band interference by reducing bleed-over limits to -57dB." The memo also says the Navy should "work with NOAA and NASA to continually assess and quantify actual impacts" and develop mitigations including "limited use of other channels, substitution of lesser-fidelity parameters, and the development of new techniques and algorithms through new research and development."

Walmart will now offer shoppers the option to have their online orders delivered the next day, following Amazon's recently announced plans to spend $800 million for one-day delivery for all Amazon Prime members. CNBC reports: Walmart said Tuesday it is rolling out next-day delivery in Phoenix, Las Vegas and Southern California over the next few days and will expand it to reach roughly 75% of American consumers by the end of 2019, including 40 of the top 50 major metros. Amazon hasn't yet detailed a timeline for its own rollout of next-day shipping. But even before its April announcement, the company had offered same-day and two-hour delivery for Prime members in certain markets for certain products and at an additional cost. Amazon's next-day shipping plan expands the number of items and ZIP codes eligible for expedited service.

Walmart isn't disclosing the cost of its latest delivery push. But the company says it has been working on it for quite some time. To start, next-day delivery will be available for about 220,000 items "most frequently purchased" online, Walmart said, including toys and electronics. The company said it plans to make more items available to ship next day over time. And the option is only free for orders over $35. Amazon, for comparison, has no minimum purchase threshold for free, next-day delivery but requires customers to have a Prime membership, which costs $119 annually.

Microsoft today is taking the unusual step of releasing security updates for unsupported but still widely-used Windows operating systems like XP and Windows 2003, citing the discovery of a "wormable" flaw that the company says could be used to fuel a fast-moving malware threat like the WannaCry ransomware attacks of 2017. From a report: The vulnerability (CVE-2019-0708) resides in the "remote desktop services" component built into supported versions of Windows, including Windows 7, Windows Server 2008 R2, and Windows Server 2008. It also is present in computers powered by Windows XP and Windows 2003, operating systems for which Microsoft long ago stopped shipping security updates. Microsoft said the company has not yet observed any evidence of attacks against the dangerous security flaw, but that it is trying to head off a serious and imminent threat.

The Verge reports on the new ad types Google announced today that will start showing up throughout its mobile products, including some that interrupt the core Google search and discovery experiences. From the report: Google searches on mobile will soon include "gallery" ads that allow advertisers to display multiple images for users to swipe through. You'll also begin to see ads in Google's discover feed -- the feed of news stories that you find built into many Android home screens, inside the Google app, and on Google's mobile homepage -- though they'll only appear in select locations for now. The new ad formats are meant to make ads a lot more noticeable. In a blog post, Google ad chief Prabhakar Raghavan says that, in tests, gallery ads resulted in "up to 25 percent more interactions" than traditional search ads.

Gallery ads will only be launching on mobile, not the desktop. Discover ads will appear in Google's mobile app, as well as on the discover feed on Android phones. Google tells us those ads won't appear in the discover feed that's built into the google.com mobile homepage. [...] The discover feed -- a personalized feed of recommended news stories that Google displays on mobile -- will also be getting ads for the first time. They'll appear just like any other story, with an image on top, a headline, and a subject field with more information. But they'll have a small badge that says "ad" to let users know it's sponsored. Those ads will extend to YouTube as well, where they'll slot in alongside recommended videos. Discover ads will also roll out later this year.

Former Google CEO Eric Schmidt has defended the company's plan to build a censored version of its search engine in China. From a report: In an interview with the BBC on Monday, Schmidt said that he wasn't involved in decisions to build the censored search platform, codenamed Dragonfly. But he insisted that there were "many benefits" to working with China and said he was an advocate of operating in the country because he believed it could "help change China to be more open." As The Intercept first revealed in August, Google developed a prototype of the censored search engine that was designed to remove content that China's ruling Communist Party regime deems sensitive. The search engine would have blacklisted thousands of words and phrases, including terms such as "human rights," "student protest," and "Nobel Prize" in Mandarin.

With employees organizing sit-ins over retaliation and continuing to agitate for change, Google is locking down internal communications. From a report: Google's top legal executive Kent Walker sent an all-staff email on Thursday informing employees that accessing documents classified as "need to know" without permission could result in termination, sources inside the company tell BuzzFeed News. After BuzzFeed News contacted Google about the email, Walker sent an update on Tuesday in the company's daily newsletter, clarifying that employees were typically only terminated when intentional violations resulted in data leaks, risks to user privacy, or harm to co-workers.

The Thursday email, titled "An important reminder on data classifications," referenced changes to Google's data security policy that were updated in October. Although the policy has been in place since 2007, and updates are visible internally, employees weren't notified by email at the time. The timing of the email announcement rattled employees who've been involved with organizing within the company's ranks and who told BuzzFeed News they saw it as a blow to internal accountability mechanisms. These employees said the "need to know" language in the data security policy leaves which particular documents are considered "need to know" up to Google's interpretation; "need to know" documents aren't necessarily labeled as such, and the punishment for accessing such documents without permission can vary, but include termination.

San Francisco, long one of the most tech-friendly and tech-savvy cities in the world, is poised to prohibit its government from using facial-recognition technology. From a report: A proposed ban is part of a broader anti-surveillance ordinance that the city's Board of Supervisors is expected to approve on Tuesday. If passed -- a majority of the board's 11 supervisors have expressed support for it -- it will make San Francisco the first city in the United States to outlaw the use of such technology by the police and other government departments. The ordinance could also spur other local governments to take similar action. Facial-recognition systems are increasingly used everywhere from police departments to rock concerts to homes, stores and schools. They are designed to identify specific people from live video feeds, recorded video footage or still photos, often by comparing their features with a set of faces (such as mugshots).

Academics have discovered a new class of vulnerabilities in Intel processors that can allow attackers to retrieve data being processed inside a CPU. From a report: The leading attack in this new vulnerability class is a security flaw named Zombieload, which is another side-channel attack in the same category as Meltdown, Spectre, and Foreshadow. Just like the first three, Zombieload is exploited by taking advantage of the speculative execution process, which is an optimization technique that Intel added to its CPUs to improve data processing speeds and performance. For more than a year, academics have been poking holes in various components of the speculative execution process, revealing ways to leak data from various CPU buffer zones and data processing operations. Meltdown, Spectre, and Foreshadow have shown how various CPU components leak data during the speculative execution process.

Today, an international team of academics -- including some of the people involved in the original Meltdown and Spectre research -- along with security researchers from Bitdefender have disclosed a new attack impacting the speculative execution process. This one is what researchers have named a Microarchitectural Data Sampling (MDS) attack, and targets a CPU's microarchitectural data structures, such as the load, store, and line fill buffers, which the CPU uses for fast reads/writes of data being processed inside the CPU. [...] In a research paper published today, academics say that all Intel CPUs released since 2011 are most likely vulnerable. Processors for desktops, laptops, and (cloud) servers are all impacted, researchers said on a special website they've set up with information about the Zombieload flaws.

Disney's takeover of Hulu is just about complete. Comcast on Tuesday agreed to sell its ownership stake in the streaming video service to Disney. The sale won't happen for at least another five years, but Disney will take full operational control of Hulu right now. From a report: The deal is a sign of how much streaming video has changed in the last few years. Hulu was at one time a joint venture between several media titans: Disney (DIS), 21st Century Fox, Comcast's NBCUniversal and Time Warner (now WarnerMedia). None of those companies had majority control. Now Hulu is effectively a Disney product. The company became the majority stakeholder in the streaming video service after it closed a deal for most of Fox's assets in March. Last month, WarnerMedia -- now owned by AT&T -- agreed to sell its 9.5% interest back to Hulu. Comcast owns roughly a third of Hulu. Under the terms of the deal, Comcast will sell its interest to Disney for Hulu's fair market value no earlier than 2024. Disney has guaranteed that the sale price will reflect a minimum total equity value of $27.5 billion for Hulu at that time, according to a press release.

Huawei is willing to sign no-spy agreements with governments, including Britain, the Chinese telecommunications company's chairman said on Tuesday as the United States pressures European countries to shun the firm over spying concerns. From a report: Washington has told allies not to use Huawei's technology to build new 5G telecommunications networks because of worries it could be a vehicle for Chinese spying, an accusation the firm has denied.

Adobe Creative Cloud subscribers who haven't updated their apps in a while may want to check their inboxes. The software company has sent out emails to customers warning them of being "at risk of potential claims of infringement by third parties" if they continue using outdated versions of CC apps, including Photoshop and Lightroom. From a report: These emails even list the old applications installed on the subscribers' systems, and in some cases, they mention what the newest available versions are. In a response to a customer complaint on Twitter, the AdobeCare account said users can only download the two most recent variants of CC apps going forward.

A spokesperson said in a statement, "Adobe recently discontinued certain older versions of Creative Cloud applications. Customers using those versions have been notified that they are no longer licensed to use them and were provided guidance on how to upgrade to the latest authorized versions." However, the spokesperson said Adobe can't comment on claims of third-party infringement, as it concerns ongoing litigation.

An anonymous reader quotes a report from The New York Times: An Israeli firm accused of supplying tools for spying on human-rights activists and journalists now faces claims that its technology can use a security hole in WhatsApp, the messaging app used by 1.5 billion people, to break into the digital communications of iPhone and Android phone users (Warning: source may be paywalled; alternative source). Security researchers said they had found so-called spyware -- designed to take advantage of the WhatsApp flaw -- that bears the characteristics of technology from the company, the NSO Group.

The spyware was used to break into the phone of a London lawyer who has been involved in lawsuits that accused the company of providing tools to hack the phones of Omar Abdulaziz, a Saudi dissident in Canada; a Qatari citizen; and a group of Mexican journalists and activists, the researchers said. There may have been other targets, they said. Digital attackers could use the vulnerability to insert malicious code and steal data from an Android phone or an iPhone simply by placing a WhatsApp call, even if the victim did not pick up the call. As WhatsApp's engineers examined the vulnerability, they concluded that it was similar to other tools from the NSO Group, because of its digital footprint. WhatsApp engineers patched the vulnerability on Monday.

"WhatsApp encourages people to upgrade to the latest version of our app, as well as keep their mobile operating system up to date, to protect against potential targeted exploits designed to compromise information stored on mobile devices," the Facebook-owned company said in a statement.

At its Accelerate 2019 event in Orlando today, Lenovo previewed "the world's first foldable PC." While we don't know the name, price tag, or ship date, we do know that the foldable PC will be part of Lenovo's flagship ThinkPad X1 line and that it will arrive in 2020. VentureBeat reports: Lenovo backs up its "the world's first foldable PC" claim by saying it looked at laptops sold by major PC manufacturers this month. None shipped more than "1 million units worldwide annually" with foldable screens. Apparently Lenovo is hoping to ship at least 1 million units of its new foldable PC in the first year.

We don't know much about the device yet, and that's on purpose. Tom Butler, Lenovo's ThinkPad marketing director, did say that the company has been working on the device for "several years" with partners Intel, Microsoft and LG. He confirmed that those three have been part of the project from the very beginning. Intel chips and Windows will be powering the foldable ThinkPad. LG is responsible for manufacturing the screen, the highlight of the device. It's a 13.3-inch single OLED 2K display with a 4:3 aspect ratio. It's also a touchscreen and will support pen input. When folded in half, the width of the device is reduced by 50%, as you might expect.

Facebook revealed last Friday that it has filed a lawsuit alleging South Korean analytics firm Rankwave abused its developer platform's data, and has refused to cooperate with a mandatory compliance audit and request to delete the data. TechCrunch reports: Facebook's lawsuit centers around Rankwave offering to help businesses build a Facebook authorization step into their apps so they can pass all the user data to Rankwave, which then analyzes biographic and behavioral traits to supply user contact info and ad targeting assistance to the business. Rankwave also apparently misused data sucked in by its own consumer app for checking your social media "influencer score." That app could pull data about your Facebook activity such as location checkins, determine that you've checked into a baseball stadium, and then Rankwave could help its clients target you with ads for baseball tickets.

The use of a seemingly fun app to slurp up user data and repurpose it for other business goals is strikingly similar to how Cambridge Analytica's personality quiz app tempted millions of users to provide data about themselves and their friends. TechCrunch has attained a copy of the lawsuit that alleges that Rankwave misused Facebook data outside of the apps where it was collected, purposefully delayed responding to a cease-and-desist order, claimed it didn't violate Facebook policy, lied about not using its apps since 2018 when they were accessed in April 2019, and then refused to comply with a mandatory audit of its data practices. Facebook Platform data is not supposed to be repurposed for other business goals, only for the developer to improve their app's user experience.

An anonymous reader quotes a report from Ars Technica: On Monday, a U.S.-based company named Hermeus announced plans to develop an aircraft that will travel at speeds of up to Mach 5. Such an aircraft would cut travel time from New York to Paris from more than 7 hours to 1.5 hours. Hermeus said it has raised an initial round of funding led by Khosla Ventures, but it declined to specify the amount. This funding will allow Hermeus to develop a propulsion demonstrator and other initial technologies needed to make its supersonic aircraft a reality, Skyler Shuford, the company's chief operating officer, told Ars.

The announcement follows three years after another company, Boom Supersonic, declared its own intentions to develop faster-than-sound aircraft. As of January 2019, Boom had raised more than $140 million toward development of its Overture airliner, envisioned to travel at Mach 2.2, which is about 10 percent faster than the Concorde traveled. Officials with Boom Supersonic have said its planes could be ready for commercial service in the mid-2020s, and they added that Virgin Group and Japan Airlines have preordered a combined 30 airplanes.

Apple's VP of Apple Pay, Jennifer Bailey, announced new NFC tags that will let iPhone users make purchases simply by tapping their phones against the stickers, without the need to download a special app first. "The company is partnering with Bird scooters, Bonobos clothing store, and PayByPhone parking meters for the initial rollout," reports 9to5Mac. From the report: Apple also announced that inside the Wallet app, users will soon be able to sign up for loyalty cards in one tap, presumably presented to users as recommendations when they make eligible purchases. Right now, physical Apple Pay transactions require bulky terminals like those you find at retail store checkouts. With the new support, an iPhone will know how to read a specially-encoded NFC tag (that can be as inert as a sticker) and automatically show the Apple Pay purchase interface when a user holds their device near it. No third-party apps or other set up required.

The obvious example is a user can ad-hoc top up their miles on a hired electric scooter simply by tapping their phone or watch to a NFC sticker on the bike. For Bonobos, it will enable simpler self-service shopping with the ability to place NFC tags directly onto clothing rails. The new Apple Pay features will be rolling out later this year, presumably with more partners onboard now that the news is public. This is yet another step towards Apple's goal of replacing the wallet.

Facebook will increase the hourly pay rate for thousands of contract workers across the United States, the company said today. From a report: Its base rate for contractors will rise from $15 an hour to $18, with slightly higher raises in cities with higher costs of living. The changes will go into effect by the middle of next year, the company said, and it will explore bringing similar raises to other sites around the world. The move comes after reporting from The Verge and others on the long-term impact of working as a contract moderator for Facebook, which has left some workers with symptoms of post-traumatic stress. Workers in larger metropolitan areas will get raises as well: from $18 to $20 an hour in Seattle and from $20 to $22 an hour in the Bay Area, New York City, and Washington, DC.

Carl Malamud believes in open access to government records, and he has spent more than a decade putting them online. You might think states would welcome the help. From a report: But when Mr. Malamud's group posted the Official Code of Georgia Annotated, the state sued for copyright infringement. Providing public access to the state's laws and related legal materials, Georgia's lawyers said, was part of a "strategy of terrorism." A federal appeals court ruled against the state, which has asked the Supreme Court to step in. On Friday, in an unusual move, Mr. Malamud's group, Public.Resource.Org, also urged the court to hear the dispute, saying that the question of who owns the law is an urgent one, as about 20 other states have claimed that parts of similar annotated codes are copyrighted.

The issue, the group said, is whether citizens can have access to "the raw materials of our democracy." The case, Georgia v. Public.Resource.Org, No. 18-1150, concerns the 54 volumes of the Official Code of Georgia Annotated, which contain state statutes and related materials. The state, through a legal publisher, makes the statutes themselves available online, and it has said it does not object to Mr. Malamud doing the same thing. But people who want to see other materials in the books, the state says, must pay the publisher.