A paper describing MineDojo, Nvidia’s generalist AI agent that can perform actions from written prompts in Minecraft, won an Outstanding Datasets and Benchmarks Paper Award at the 2022 NeurIPS (Neural Information Processing Systems) conference, Nvidia revealed on Monday.
To train the MineDojo framework to play Minecraft, researchers fed it 730,000 Minecraft YouTube videos (with more than 2.2 billion words transcribed), 7,000 scraped webpages from the Minecraft wiki, and 340,000 Reddit posts and 6.6 million Reddit comments describing Minecraft gameplay.
From this data, the researchers created a custom transformer model called MineCLIP that associates video clips with specific in-game Minecraft activities. As a result, someone can tell a MineDojo agent what to do in the game using high-level natural language, such as “find a desert pyramid” or “build a nether portal and enter it,” and MineDojo will execute the series of steps necessary to make it happen in the game.
MineDojo aims to create a flexible agent that can generalize learned actions and apply them to different behaviors in the game. As Nvidia writes, “While researchers have long trained autonomous AI agents in video-game environments such as StarCraft, Dota, and Go, these agents are usually specialists in only a few tasks. So Nvidia researchers turned to Minecraft, the world’s most popular game, to develop a scalable training framework for a generalist agent—one that can successfully execute a wide variety of open-ended tasks.”
The award-winning paper, “MINEDOJO: Building Open-Ended Embodied Agents with Internet-Scale Knowledge,” debuted in June. Its authors include Linxi Fan of Nvidia and Guanzhi Wang, Yunfan Jiang, Ajay Mandlekar, Yuncong Yang, Haoyi Zhu, Andrew Tang, De-An Huang, Yuke Zhu, and Anima Anandkumar of various academic institutions.
You can see examples of MineDojo in action on its official website, and the code for MineDojo and MineCLIP is available on GitHub.
Microsoft said on Friday that an Iranian nation-state group already sanctioned by the US government was behind an attack last month that targeted the satirical French magazine Charlie Hebdo and thousands of its readers.
The attack came to light on January 4, when a previously unknown group calling itself Holy Souls took to the Internet to claim it had obtained a Charlie Hebdo database that contained personal information for 230,000 of its customers. The post said the database was available for sale at the price of 20 BTC, or roughly $340,000 at the time. The group also released a sample of the data that included the full names, telephone numbers, and home and email addresses of people who had subscribed to, or purchased merchandise from, the publication. French media confirmed the veracity of the leaked data.
The release of the sample put the customers at risk of online targeting or physical violence by extremist groups, which have retaliated against Charlie Hebdo in recent years for its satirical treatment of matters pertaining to the Muslim religion and Islamic countries such as Iran. The retaliation included the 2015 shooting by two French Muslim terrorists and brothers at Charlie Hebdo offices that killed 12 and injured 11 others. To further gin up attention to the breached data, a flurry of fake personas—one falsely claiming to be a Charlie Hebdo editor—took to social media to discuss and publicize the leak.
On Friday, Clint Watts, the general manager of Microsoft’s Digital Threat Analysis Center, wrote:
We believe this attack is a response by the Iranian government to a cartoon contest conducted by Charlie Hebdo. One month before Holy Souls conducted its attack, the magazine announced it would be holding an international competition for cartoons “ridiculing” Iranian Supreme Leader Ali Khamenei. The issue featuring the winning cartoons was to be published in early January, timed to coincide with the eighth anniversary of an attack by two al-Qa’ida in the Arabian Peninsula (AQAP)-inspired assailants on the magazine’s offices.
The tactics, techniques, and procedures of the influence campaign led Microsoft researchers to conclude it was the work of Emennet Pasargad, an Iranian group that has long been monitored and targeted by the US government. The FBI said in January 2022 that Emennet Pasargad was behind “a multi-faceted campaign to interfere in the 2020 US presidential election.”
Participants in the operation obtained confidential US voter information from at least one state election website, sent threatening emails designed to intimidate voters, and published a video airing disinformation concerning non-existent voting vulnerabilities. The group also claimed affiliation with the neo-fascist group Proud Boys to further intimidate voters.
Last October, the FBI said that Emennet Pasargad targeted groups in Israel with “cyber-enabled information operations that included an initial intrusion, theft, and subsequent leak of data, followed by amplification through social media and online forums, and in some cases the deployment of destructive encryption malware.”
The US Treasury in 2021 placed sanctions on Emennet Pasargad and six Iranian nationals who are members, citing their attempts “to sow discord and undermine voters’ faith in the US electoral process.”
Friday’s post said Microsoft had “high confidence” that the group, which the company refers to as Neptunium, was behind the Charlie Hebdo influence campaign. The assessment was based on elements including:
A hacktivist persona claiming credit for the cyberattack
Claims of a successful website defacement
Leaking of private data online
The use of inauthentic social media “sockpuppet” personas—social media accounts using fictitious or stolen identities to obfuscate the account’s real owner for the purpose of deception—claiming to be from the country that the hack targeted to promote the cyberattack using language with errors obvious to native speakers
Impersonation of authoritative sources
Contacting news meida organizations
Microsoft said the January campaign used French-language sockpuppet social media accounts, many with low follower counts, to amplify the leak and “distribute antagonistic messaging.” The accounts also posted criticisms of the cartoon competition aimed at Khamenei.
“Crucially, before there had been any substantial reporting on the purported cyberattack, these accounts posted identical screenshots of a defaced website that included the French-language message: ‘Charlie Hebdo a été piraté’ (‘Charlie Hebdo was hacked’),” Watts wrote.
Shortly after that, at least two social media accounts—one purporting to belong to a tech executive and the other to a Charlie Hebdo editor—posted screenshots of the leaked customer data.
The campaign Microsoft has documented is the latest reminder that social media is often manipulated by special interest groups—some with deep pockets. People would do well to remember this manipulation and be careful to verify claims before spreading them further.
If your main problem with the Microsoft Store is that you get too many relevant results when you search for apps, good news: Microsoft is officially launching Microsoft Store Ads, a way for developers to pay to get their apps in front of your eyes when you go to the store to look for something else.
Microsoft’s landing page for the feature says the apps will appear during searches and in the Apps and Gaming tabs within the app. Developers will be able to track whether and where users see the ads and whether they’re downloading and opening the apps once they see the ads.
Microsoft also provided an update on the health of the Microsoft Store, pointing to 2022 as “a record year,” with more than 900 million unique users worldwide and “a 122% year-over-year increase in developer submissions of new apps and games.” Microsoft has steadily loosened its restrictions on Store apps in the last year or two, allowing in traditional Win32 apps and also leaning on Amazon’s Android app store and the Windows Subsystem for Android to expand its selection.
The company launched a “pilot program” of the Microsoft Store Ads back in September of 2022, and the look of the ads doesn’t appear to have changed much since then. Ads will be served to Microsoft Store users on Windows 10 and Windows 11 and are only available to developers who have already published their apps to the store.
These kinds of ads are usually described in benign terms—that they’re merely a way for the developers on the Microsoft Store to boost their work and find more users. The reality is that similar ads on Apple’s platforms, at least in my experience, tend to be either irrelevant (ads for Twitter or Truth Social on a search for Mastodon clients), annoying (ads for shovelware free-to-play games), actively malicious (the brief period where gambling ads took over the store), or some combination of all three.
The new addition may or may not turn up relevant search results for users, but it does add more advertisements to a platform that already has plenty of them. A fresh Windows 11 install from a USB stick automatically pulls down a range of third-party apps from the Microsoft Store the first time you connect to the Internet, and Windows includes plenty of Microsoft house ads for Edge, Bing, Microsoft Start, Microsoft 365, OneDrive, and other products and features.
Searching Google for downloads of popular software has always come with risks, but over the past few months, it has been downright dangerous, according to researchers and a pseudorandom collection of queries.
“Threat researchers are used to seeing a moderate flow of malvertising via Google Ads,” volunteers at Spamhaus wrote on Thursday. “However, over the past few days, researchers have witnessed a massive spike affecting numerous famous brands, with multiple malware being utilized. This is not ‘the norm.’”
One of many new threats: MalVirt
The surge is coming from numerous malware families, including AuroraStealer, IcedID, Meta Stealer, RedLine Stealer, Vidar, Formbook, and XLoader. In the past, these families typically relied on phishing and malicious spam that attached Microsoft Word documents with booby-trapped macros. Over the past month, Google Ads has become the go-to place for criminals to spread their malicious wares that are disguised as legitimate downloads by impersonating brands such as Adobe Reader, Gimp, Microsoft Teams, OBS, Slack, Tor, and Thunderbird.
On the same day that Spamhaus published its report, researchers from security firm Sentinel One documented an advanced Google malvertising campaign pushing multiple malicious loaders implemented in .NET. Sentinel One has dubbed these loaders MalVirt. At the moment, the MalVirt loaders are being used to distribute malware most commonly known as XLoader, available for both Windows and macOS. XLoader is a successor to malware also known as Formbook. Threat actors use XLoader to steal contacts data and other sensitive data from infected devices.
The MalVirt loaders use obfuscated virtualization to evade end-point protection and analysis. To disguise real C2 traffic and evade network detections, MalVirt beacons to decoy command and control servers hosted at providers including Azure, Tucows, Choopa, and Namecheap. Sentinel One researcher Tom Hegel wrote:
As a response to Microsoft blocking Office macros by default in documents from the Internet, threat actors have turned to alternative malware distribution methods—most recently, malvertising. The MalVirt loaders we observed demonstrate just how much effort threat actors are investing in evading detection and thwarting analysis.
Malware of the Formbook family is a highly capable infostealer that is deployed through the application of a significant amount of anti-analysis and anti-detection techniques by the MalVirt loaders. Traditionally distributed as an attachment to phishing emails, we assess that threat actors distributing this malware are likely joining the malvertising trend.
Given the massive size of the audience threat actors can reach through malvertising, we expect malware to continue being distributed using this method.
Google representatives declined an interview. Instead, they provided the following statement:
Bad actors often employ sophisticated measures to conceal their identities and evade our policies and enforcement. To combat this over the past few years, we’ve launched new certification policies, ramped up advertiser verification, and increased our capacity to detect and prevent coordinated scams. We are aware of the recent uptick in fraudulent ad activity. Addressing it is a critical priority and we are working to resolve these incidents as quickly as possible.
Anecdotal evidence that Google malvertising is out of control isn’t hard to come by. Searches seeking software downloads are probably the most likely to turn up malvertising. Take, for instance, the results Google returned for a search Thursday looking for “visual studio download”:
Clicking that Google-sponsored link redirected me to downloadstudio[.]net, which is flagged by VirusTotal as malicious by only a single endpoint provider:
On Thursday evening, the download this site offered was detected as malicious by 43 antimalware engines: