The bizarre release of sensitive US government materials soon after their creation signals a potential shift to near-real-time unauthorized disclosures.

the United States government has been scrambling this week, thanks to a leak of sensitive Pentagon documents that include an array of recent intelligence updates and Joint Chiefs of Staff briefings. Notably, the documents seem to be very recent, dating from as early as January to early March. The trove was first posted to Discord a few weeks ago, before some of the documents got pickup more recently on Russian Telegram channels and then Twitter.

Initial reporting about the situation—including by _The New York Times_, which broke the story—has focused on roughly 100 documents, but some reports indicate that even more secret documents may have been shared on Discord over the past few months. The documents are photographs of printed-out presentation slides. Some of the papers had been folded and unfolded before being photographed, and some of the photos capture slivers of other objects that were on the desk with the papers.

Researchers say the leak ranks high among other prominent recent revelations about clandestine US government activity—a list that includes information from Edward Snowden about the NSA's bulk surveillance activity, details of the CIA's hacking capabilities in the Vault7 revelations published by WikiLeaks, and NSA hacking tools revealed in the Shadow Brokers leak. But this latest leak has some specific characteristics reflective of the current moment: It is relatively small and contains fresh information rather than a large trove of months- or years-old data. And while it is not yet clear who leaked the documents or what their motivation was, initial indications from Discord activity suggest that the leaker may have been trying to show off to their gaming friends, and might even be a teenager or young adult.

“I am intrigued by the idea of a small, pinpoint leak phenomenon,” says Dan Meyer, a partner at the law firm Tully Rinckey who works on federal employment and national security matters. Meyer was formerly a federal investigator and whistle-blowing expert within the US government. The use of “strategic leaks” has been a tactic of top officials “for a very long time,” Meyer says. “But now the technology issue is very real with phones and the ability to move these documents in ways the government didn’t anticipate.”

The latest leaked documents reveal details about the war in Ukraine, including information about the Ukrainian army's air defenses and plans for a future counteroffensive against Russia. Notably, the trove also reveals details about Russia's war effort and exposes the degree to which the United States intelligence community has penetrated Russia's military and intelligence services. 

As with any leak of state secrets, the most impactful revelations long-term generally relate to methods and access rather than specific information. That means Russian intelligence, for example, will likely make changes to its operations in response to the relvations to evade American operators. The leak also includes indications that the US has been spying on allies like South Korean and Israeli officials, a move that is not necessarily surprising but diplomatically embarrassing.

“We're still investigating how this happened, as well as the scope of the issue,” Chris Meagher, the US assistant to the secretary of defense for public affairs, told reporters on Tuesday. “There have been steps to take a closer look at how this type of information is distributed and to whom. We are also still trying to assess what might be out there.”

Analysts emphasize that the scale of the leak may be much larger than what has emerged publicly so far, and that the situation reflects concerning lapses in the Pentagon's systems for handling secrets.

“It seems like the Department of Defense thought they had sufficient controls in place to detect would-be leakers after incidents like Snowden,” says Jake Williams, a former NSA hacker and an analyst with IANS Research, a cybersecurity consultancy firm. “But obviously, whoever is doing this got around that or learned from past techniques and mistakes.”

As real-time accounts of everything from wars to natural disasters play out on social media and other digital communication platforms, it makes sense that leaking has increasingly become targeted and agile as well. So-called “hack and leak” operations have demonstrated this in recent years, with, for example, state-backed attackers leaking excerpts of government officials' digital communications or strategy documents from political campaigns. Hactivists have also increasingly used precision leaks around the world. Ukrainian advocates have, for example, repeatedly doxed Russian military officials and intelligence agents. And they've leaked data stolen from Russian government agencies and private companies since the Kremlin's full-scale invasion of Ukraine in February 2022. 

For now, as the US government aggressively investigates the latest Pentagon leaker and their motives, Meyer says his advice for US federal employees, contractors, or anyone with a US security clearance is to mind their business.

“Just because you have the ‘need to know,’ that doesn't mean you have the need to go look at it,” he says. “That person should not access these documents anywhere on the web. This stuff has now been identified as classified, and even if it's well intentioned, if you view classified information in a non-SCIF environment, you've committed a technical infraction. It’s a mess, don’t do it.”

Leaked Pentagon Documents May Herald a New Era of Revelations

To beat back fake accounts, the professional social network is rolling out new tools to prove you work where you say you do and are who you say you are.

In the never-ending battle against online impersonation scams, the professional social media platform LinkedIn announced today a set of new verification features that enable users to authenticate aspects of their identities and job histories. Crucially, users will now have a few different options to verify their identity and current jobs on LinkedIn. That way, if someone tries to make a copycat LinkedIn account, there can be clear differences between the imposter account and the verified profile.  

LinkedIn facilitates verification in three ways that are all free to individual users. The most low-key option launching today is to verify your current employer by receiving a security code on your work email and entering it into LinkedIn. The social media platform has recently been piloting this work email verification feature with a small group of companies.

The second option is to verify your identity on LinkedIn through the airport security service Clear. The authentication company will take your United States phone number and government-issued ID and use the information to verify your name. You have to weigh whether you want to trust a third party like Clear with your personal data, but the option might be particularly appealing if you already use the company for travel verification and they have your data on file anyway.

The third verification feature allows users to confirm their name and current employer through the Microsoft Entra Verified ID credential, a workplace identification platform Microsoft launched last year. This option will have a slower rollout, and it will be available at the end of the month to employees at a few dozen pilot companies that are already enrolled in Entra.

Once you add any of these verifications to your LinkedIn account, a new Verification field will show on your profile with details.

“Through all these new, free features, we’re helping give you the confidence that who you’re connecting with and the content you come across is trusted and authentic,” Oscar Rodriguez, LinkedIn’s vice president of product management, wrote in a blog post today.

After noting in June 2022 that the company had seen “a rise in fraudulent activity” on its platform and across the web, LinkedIn announced efforts in October to detect and remove more fake accounts, expand verification, and generally “boost authenticity” for its more than 900 million users. Today’s announcements dramatically expand the scale and scope of those verification initiatives.

Having the ability to verify components of your identity and employment won't keep attackers from generating fictional personas and even fake companies to "verify" phony jobs. But if job verification is widely adopted on LinkedIn, it will make it harder for bad actors to impersonate legitimate accounts and build out compelling fake personas.

“By simply looking for a Verification, members and organizations can be more confident that the people they collaborate with are authentic and that work affiliations on their profiles are accurate," Microsoft wrote in a blog post published today.

The rollout of the three verification options seems to be starting in the US. Rodriguez wrote in the blog post that not all LinkedIn users around the world will have access to the verification methods immediately. “We will expand availability and ways for you to participate over time,” he wrote.

Microsoft notes that its Verified ID tools are based on open standards, which the company says will make it easier for the LinkedIn system to interoperate with existing employee management or HR systems. Microsoft has been investing in and developing decentralized identity schemes for a few years. Bringing the technology to nearly a billion LinkedIn users will be a big step toward wider adoption.

“On LinkedIn, members will see an option to verify their workplace on their profile,” Microsoft wrote in its blog post. “With a few taps on their phone, members can get their digital employee ID from their organization and choose to share it on LinkedIn.”

Researchers who monitor digital crime say that robust verification methods really can reduce the prevalence of online scams—or at least make attackers' jobs more difficult in consequential ways. Since its acquisition by Elon Musk last year, for example, Twitter has been criticized for transitioning its verification model to a service that anyone with a credit card can pay for.

“Many actors will go make a LinkedIn profile pretending, say, I’m from WIRED, and I’m a director. They can arbitrarily put whatever they want,” says Ronnie Tokazowski, principal threat adviser at the cybersecurity firm Cofense. “There are cases where scammers reach out to victims through LinkedIn and start a conversation for investment scams or to distribute fake job postings. Fraud cases definitely happen through LinkedIn.”

While the professional social network is far from the only platform where criminals and nation-backed actors are hustling, LinkedIn still retains a particular air of legitimacy and the sense that a professional social network is benign and mundane. It may therefore be a platform where providing options for information verification is particularly urgent.

LinkedIn Verification Now Lets You Verify Your Job and Account

Your iPhone, iPad, and Mac now have a built-in password feature, complete with two-factor authentication.

Most people don't use a password manager or two-factor authentication—even people who know it's a good idea—because installing and managing _yet another app_ just sounds exhausting. Well, if you're an Apple user, you don't need another app anymore: Your device can manage your passwords and generate two-factor authentication codes for you, and you can even sync them with a Windows computer.

Password managers are important. Why? To quickly summarize, using the same password for every website and app is an open invitation for hackers to access all of your accounts. That's because passwords regularly leak, and a leaked password on one site can give hackers access to all your other accounts if you use the same password everywhere. It's best, then, to use a totally different password on every site, but no human being can remember that many passwords.

Password managers are the best solution we have at the moment, because they can generate, and then store, secure passwords for all of your services. Most people don't use one, though, because such apps can be complicated to learn, and the best ones aren't free.

So it's great that Apple offers such functionality. But there's a downside: It's a little buried, if not outright hidden. Still, if you're a Safari user with multiple Apple devices, this feature means you can quickly generate and save secure passwords for all of your accounts. Here's how. Note that you'll need a (free) iCloud account for this service to sync passwords between devices, though if you're an Apple user you almost certainly already have one.

Create Passwords

To get started, you may need to enable the feature, which you can find in System Settings on your device under **Passwords**. Make sure **iCloud Keychain** is turned on. Windows users should also install iCloud for Windows, which can sync your passwords with Chrome or Microsoft Edge.

The simplest way to add passwords to Apple's hidden password manager is to just start using your devices and saving passwords as you go. When you sign into any online account in Safari, or in any app on your iPhone or iPad, there's generally a pop-up asking if you want to save the password in your **iCloud Keychain for AutoFill.** This is the simplest way to add accounts: Just hit the **Save Password** button and your username and password will be saved.

Alternatively, if you're signing up for a new account, you will generally be offered an automatically generated strong password—if not, you can tap the key icon at the top of the keyboard on mobile or in the right side of the password field on desktop. Either way you should see the **Add New Password** option, which can automatically create a strong password for you.

From now on, when you log in to the site, your device will offer to fill out the username and password for you. It will generally use TouchID, FaceID, or your system password to confirm you identity, after which the username and password field will be filled in. This saves you from having to remember the passwords, and even the usernames, you use to log in to websites.

Browse and Edit Your Passwords

You might be wondering _where_, exactly, all of these passwords are saved. Let's head back to **Passwords** in System Settings. Here you will find a list of all the passwords you've saved.

Apple via Justin Pot

At the top of the list is a **Security Recommendations** function, which will cross-reference your saved passwords with known lists of leaked accounts. This is a useful way to know if any of your passwords absolutely need to be changed.

Below that is a search bar, which you can use to quickly find any account. Open an account to see the username, password, and URLs associated with the account. You can also add a note to any account, if you want.

Two-Factor Authentication

We've talked about how two-factor authentication keeps you more secure, but basically it means that a hacker who gets your password won't be able to log in unless they _also_ have physical access to your device. Generally two-factor authentication requires installing _yet another_ app, for generating codes, but Apple devices have this feature built in, and they can even fill in the field for you.

Head back to your list of passwords in the System Settings app. Open any account and you'll see a **Set Up Verification Code** field.

Apple via Justin Pot

Tap this and you can set up two-factor authentication for the application. How to do this depends on the specific service you're setting it up for, but it's generally in the settings of the specific app or website. You'll have a QR code to sign or, alternatively, a long code to copy.

Apple via Justin Pot

After setting this up, your Apple device will automatically offer verification codes for you every time you log in to the service on any of your devices. It's really slick, and it's a lot faster than applications like Authy or Google Authenticator.

Importing and Exporting Passwords

Note that if you have an existing password manager, you can import your passwords to Apple's system. Head back to **Passwords** in the settings app and hit the three-dot button on the right above your list of passwords. Here you will see the option to **Import passwords**.

Apple via Justin Pot

You will need to export your passwords to a CSV file before you can use this functionality. Here are instructions for the leading password managers:

*   1Password
*   LastPass
*   Bitwarden
*   Dashlane
*   Google Chrome

You can also **Export** your Apple passwords to a CSV file from here, allowing you to import them into one of these password managers. We've outlined  the best password managers for you; my personal recommendations are Bitwarden, which is free and open source, and 1Password, which is powerful and free to use.

There's not a lot of reason for the Apple faithful to do this, though. As we've outlined, Apple's password system does most of what these applications can do. The main problem is that they're hidden. 

Recently the blogger Cabel Sasser argued that Apple Passwords needs an app, which is part of why I wrote this guide. I'm not sure if I agree with his premise—I suspect most people would simply ignore any “Passwords” app, the way they ignore most applications that come bundled with their devices. Still, it is true that all of this functionality is pretty buried. I hope this article helps you find it.

How to Use Apple’s New All-In-One Password Manager

Plus: 119 arrested during a sting on the Genesis dark-web market, the IRS aims to buy an online mass surveillance tool, and more.

A United States Immigration and Customs Enforcement database WIRED obtained through a Freedom of Information Act request shows that the agency has been leaning on a certain type of administrative subpoena to collect data from elementary schools, abortion clinics, and other vulnerable populations. And new details about a recent supply chain attack against the VoIP software 3CX indicate that attackers—likely hackers working for the North Korean government—were targeting cryptocurrency companies in the broad assault.

We also looked at this week’s move by Italy’s data regulator, Garante per la Protezione dei Dati Personali, to temporarily stop OpenAI from incorporating Italians’ personal information into training data. In response, the company has currently stopped people in Italy from accessing its generative AI platform, ChatGPT. Meanwhile, we explored the dangerous missing security defense in the US agriculture sector and the nation’s food supply chain, and we went deep on the saga of a small US gadget blog that found troubling flaws in foreign security cameras and took on the Chinese surveillance industry to get them fixed.

In virtual private network news, the open source VPN Amnezia has been allowing users in Russia to stay one step ahead of the Kremlin’s inveterate censorship and digital control. And the Tor Project collaborated with the open source VPN maker Mullvad to create a new privacy-focused browser that incorporates the VPN of your choosing.

Plus, there’s more. Each week, we round up the security news we didn’t cover in-depth ourselves. Click the headlines to read the full stories, and stay safe out there.

The Chinese ecommerce giant Pinduoduo has more than 750 million customers a month and sells a vast array of products and groceries. But cybersecurity researchers who analyzed the company’s Android app found that it is laced with invasive malware that exploits Android vulnerabilities to take control of users’ devices—gaining access to data from other apps, changing system settings, and monitoring people’s digital activity in a number of ways. 

Current and former Pinduoduo employees told CNN that the company has a specific initiative to discover Android vulnerabilities and develop exploits. The goal is allegedly to increase sales by monitoring customers and competitors. CNN said there is no specific evidence that Pinduoduo gives the data it steals to Beijing, but under Chinese law that would be very possible. Google suspended the app from its Play Store in late March, but the app store is banned in China, so Android users typically download their apps from local app stores anyway. In the past, Pinduoduo has rejected “the speculation and accusation that \[the\] Pinduoduo app is malicious,” but it did not respond to multiple CNN requests for comment on the new findings. Tech giants around the world are often criticized for their massive, even excessive data collection practices. But researchers said that Pinduoduo’s app was particularly egregious.

Law enforcement from 17 counties collaborated on the takedown this week of the widely used digital criminal marketplace Genesis, known for hawking massive quantities of stolen login credentials and access tokens. Police seized the site’s infrastructure and also executed a massive campaign in multiple countries to conduct 208 property searches and arrest 119 of the site’s alleged users. The FBI and Dutch National Police led the effort with support from Europol and many others. “Working across 45 of our FBI Field Offices and alongside our international partners, the Justice Department has launched an unprecedented takedown of a major criminal marketplace that enabled cybercriminals to victimize individuals, businesses, and governments around the world,” US attorney general Merrick Garland said in a statement. “Our seizure of Genesis Market should serve as a warning to cybercriminals who operate or use these criminal marketplaces.”

Just in time for tax day, public procurement records reviewed by Motherboard show that the US Internal Revenue Service is interested in purchasing an internet surveillance tool from Team Cymru, a company that makes digital monitoring products. The FBI and US military are already customers. The tool gives users access to “netflow” data, which reveals broad internet activity, including interactions like server communication. Without such surveillance tools, only a server’s host or operator and internet service provider would have access to such data. The records also indicate that the IRS is looking to purchase access to a number of cybersecurity products for defense.

Tesla vehicles incorporate a number of cameras, but the video they capture is supposed to be locked down so you have privacy in your own car. However, Reuters found that Tesla employees shared embarrassing and “highly invasive” videos and images from customers’ cars on an internal company communication platform between 2019 and 2022. Some of the footage was simply of dogs or comical road signs, but it also captured an array of compromising situations, including nudity. Tesla didn’t respond to detailed questions from Reuters about the findings.

The Chinese spy balloon that caused an uproar as it floated over the US early this year made multiple passes over sensitive military sites and successfully collected some electronic signals, like those from communications and weapons systems, according to three current and former officials who spoke to NBC News. The US government had said at the time that it was taking steps to block the balloon from collecting anything useful. The three officials added, though, that the US’s countermeasures succeeded at substantially reducing the amount of information the balloon was able to collect.

Pinduoduo, a Top Chinese Shopping App, Is Laced With Malware

Amnezia, a free virtual private network, allows users to set up their own servers, making it harder for Moscow to block this portal to the outside world.

The Russian government has banned more than 10,000 websites for content about the war in Ukraine since Moscow launched the full-scale invasion in February 2022. The blacklist includes Facebook, Twitter, Instagram, and independent news outlets. Over the past year, Russians living inside the country have turned to censorship circumvention tools such as VPNs to pierce through the information blockade.

But as dozens of virtual private networks get blocked, leaving users scrambling to maintain their access to free information, local activists and developers are coming up with new solutions. One of them is Amnezia VPN, a free, open source VPN client.

“We even do not advertise and promote it, and new users are still coming by the hundreds every day,” says Mazay Banzaev, Amnezia VPN’s founder.

Unlike commercial VPNs that route users through company servers, which can be blocked, Amnezia VPN makes it simple for users to buy and set up their own servers. This allows them to choose their own IP address and use protocols that are harder to block.

“More than half of the commercial VPNs in Russia have been blocked because it’s easy enough to block them: They do not block them by protocols, but by IP addresses,” says Banzaev. “\[Amnezia\] is an order of magnitude more resilient than a typical commercial VPN.”

Amnezia VPN is similar to Outline, a free and open source tool developed by Jigsaw, a subsidiary of Google. Amnezia was created in 2020 during a hackathon supported by Russian digital rights organization Roskomsvoboda. Even then, “it was clear that things were moving toward stricter censorship,” says Banzaev.

Russian authorities have been attempting to control tools such as VPNs and anonymous proxy servers for years, including by introducing a law regulating these tools in 2017. Since Russia's invasion of Ukraine, however, the Kremlin has escalated its efforts to control information. 

Just days after Russian troops headed toward Kyiv, Ukraine’s capital, Vladimir Putin signed legislation that criminalizes spreading "fake" information about the war, with a penalty of up to 15 years in prison. Most independent news outlets are now blocked, with editors and journalists ending up in prison, leaving Russians with state propaganda.

This has made VPNs and other censorship circumvention tools all the more important, says Stanislav Shakirov, cofounder of Roskomsvoboda and founder of tech development organization Privacy Accelerator. “If internet users in Russia stop receiving information other than state information,” he says, “we will have no hope of any processes leading to a change in the current regime.”

The Kremlin is, of course, not giving up on its crackdown. In September 2022, Roskomnadzor, the main government body responsible for internet censorship, announced it would block six popular VPN services, including ExpressVPN and NordVPN. This was followed in March 2023 by announcements that VPNs refusing to provide data to domestic intelligence agencies would be blocked in Russia, as well as proposals to restrict anonymization tools such as virtual phone numbers. Messaging app Telegram, which saw a steep rise in popularity in Russia after the invasion, has been offering virtual phone numbers since December 2022.

Free VPN Amnezia Helps Users Avoid Censorship in Russia

Without an information sharing and analysis center, the country’s food and agriculture sector is uniquely vulnerable to hackers.

Just-in-time logistics mean that even short-term cyberattacks can have serious consequences. Hacks that disrupt fertilizer or pesticide production can force farmers to sit out planting seasons. Breaches at meat-packing plants can cause destabilizing supply shortages. Tampering at a food processing firm can lead to deadly contamination. Already, ransomware attacks that have forced companies to shut down operations for a week have left schools without milk, juice, and eggs, according to Sachs.

“A major disruption in this sector leads to immediate public health and safety issues,” says Mark Montgomery, who served as executive director of the Cyberspace Solarium Commission.

Despite being increasingly vulnerable, Sachs says, the food and agriculture sector still “doesn’t really understand the threat mindset” as well as higher-profile sectors, like financial services and energy, do.

Critical Businesses, Limited Support

Today, food and agriculture is one of four critical infrastructure sectors (out of 16) without an ISAC, along with dams, government facilities, and nuclear reactors and materials.

The food and agriculture sector was one of the first to launch such a center, in 2002, but it disbanded in 2008 because few companies were sharing information through it. Members were afraid that such openness jeopardized their competitive advantages and exposed them to regulatory action. Now, Sachs says, businesses worry that exchanging information with each other could prompt antitrust lawsuits, even though such collaboration is legal.

Some companies participate in a Food and Agriculture Special Interest Group (SIG) housed inside the IT-ISAC, which gives them access to data and analysis from some of the world’s biggest tech companies, as well as resources like playbooks for confronting specific hacker groups.

“Our work with the industry has really expanded over the last three years or so,” says IT-ISAC executive director Scott Algeier. In that same time period, the IT-ISAC has recorded 300 ransomware attacks on the food and agriculture sector.

But the SIG’s offerings are limited, Sachs argues. It doesn’t hold regular large-scale exercises simulating attacks on food and agriculture firms, doesn’t staff a 24/7 watch center that constantly monitors these firms’ infrastructure (along with related events like severe weather and supply chain disruptions), and can’t automatically generate insights and alerts by comparing classified government intelligence with data from sensors inside that infrastructure. “I appreciate everything Scott is doing over there,” Sachs says. “It's a very good thing. But it's not an ISAC.”

Algeier says the IT-ISAC has hosted exercises focused on the food and agriculture sector and that “members can reach out to us 24/7 if needed.”

But the sector needs its own ISAC that can “analyze the threat and provide a true operational assessment,” says Brian Harrell, a former assistant director for infrastructure security at the US Cybersecurity and Infrastructure Security Agency (CISA).

Pfluger says, “Plenty of folks I’ve spoken with think there needs to be a dedicated ISAC.”

Companies also need more support from the federal government.

The US Department of Agriculture, the industry’s sector risk management agency, is “significantly less effective” than other SRMAs, Montgomery says. The USDA doesn’t even have dedicated funding for its security support, which includes biannual sector-wide meetings, weekly threat bulletins, and occasional town halls.

The Dangerous Weak Link in the US Food Chain

Italy’s recent ban of Open AI’s generative text tool may just be the beginning of ChatGPT's regulatory woes.

When OpenAI released GPT-3 in July 2020, it offered a glimpse of the data used to train the large language model. Millions of pages scraped from the web, Reddit posts, books, and more are used to create the generative text system, according to a technical paper. Scooped up in this data is some of the personal information you share about yourself online. This data is now getting OpenAI into trouble. 

On March 31, Italy’s data regulator issued a temporary emergency decision demanding OpenAI stop using the personal information of millions of Italians that’s included in its training data. According to the regulator, Garante per la Protezione dei Dati Personali, OpenAI doesn’t have the legal right to use people’s personal information in ChatGPT. In response, OpenAI has stopped people in Italy from accessing its chatbot while it provides responses to the officials, who are investigating further. 

The action is the first taken against ChatGPT by a Western regulator and highlights privacy tensions around the creation of giant generative AI models, which are often trained on vast swathes of internet data. Just as artists and media companies have complained that generative AI developers have used their work without permission, the data regulator is now saying the same for people’s personal information.

Similar decisions could follow all across Europe. In the days since Italy announced its probe, data regulators in France, Germany, and Ireland have contacted the Garante to ask for more information on its findings. “If the business model has just been to scrape the internet for whatever you could find, then there might be a really significant issue here,” says Tobias Judin, the head of international at Norway’s data protection authority, which is monitoring developments. Judin adds that if a model is built on data that may be unlawfully collected, it raises questions about whether anyone can use the tools legally.

Italy’s blow to OpenAI also comes as scrutiny of large AI models is steadily increasing. On March 29, tech leaders called for a pause on the development of systems like ChatGPT, fearing its future implications. Judin says the Italian decision highlights more immediate concerns. “Essentially, we’re seeing that AI development to date could potentially have a massive shortcoming,” Judin says.

The Italian Job

Europe’s GDPR rules, which cover the way organizations collect, store, and use people’s personal data, protect the data of more than 400 million people across the continent. This personal data can be anything from a person’s name to their IP address—if it can be used to identify someone, it can count as their personal information. Unlike the patchwork of state-level privacy rules in the United States, GDPR’s protections apply if people’s information is freely available online. In short: Just because someone’s information is public doesn’t mean you can vaccuum it up and do anything you want with it.

Italy’s Garante believes ChatGPT has four problems under GDPR: OpenAI doesn’t have age controls to stop people under the age of 13 from using the text generation system; it can provide information about people that isn’t accurate; and people haven’t been told their data was collected. Perhaps most importantly, its fourth argument claims there is “no legal basis” for collecting people’s personal information in the massive swells of data used to train ChatGPT.

“The Italians have called their bluff,” says Lilian Edwards, a professor of law, innovation, and society at Newcastle University in the UK. “It did seem pretty evident in the EU that this was a breach of data protection law.”

Broadly speaking, for a company to collect and use people’s information under GDPR, they must rely on one of six legal justifications, ranging from someone giving their permission to the information being required as part of a contract. Edwards says that in this instance, there are essentially two options: getting people’s consent—which OpenAI didn’t do—or arguing it has “legitimate interests” to use people’s data, which is “very hard” to do, Edwards says. The Garante tells WIRED it believes this defense is “inadequate.”

OpenAI’s privacy policy doesn’t directly mention its legal reasons for using people’s personal information in training data but says it relies upon “legitimate interests” when it “develops” its services. The company did not respond to WIRED’s request for comment. Unlike with GPT-3, OpenAI has not publicized any details of the training data that went into ChatGPT, and GPT-4 is thought to be several times larger.

However, GPT-4’s technical paper includes a section on privacy, which says its training data may include “publicly available personal information,” which comes from a number of sources. The paper says OpenAI takes steps to protect people’s privacy, including “fine-tuning” models to stop people asking for personal information and removing people’s information from training data “where feasible.”

“How to collect data lawfully for training data sets for use in everything from just regular algorithms to some really sophisticated AI is a critical issue that needs to be solved now, as we’re kind of on the tipping point for this sort of technology taking over,” says Jessica Lee, a partner at law firm Loeb and Loeb.

The action from the Italian regulator—which is also taking on the Replika chatbot—has the potential to be the first of many cases examining OpenAI’s data practices. GDPR allows companies with a base in Europe to nominate one country that will deal with all of its complaints—Ireland deals with Google, Twitter, and Meta, for instance. However, OpenAI doesn’t have a base in Europe, meaning that under GDPR, every individual country can open complaints against it. 

Model Data

OpenAI isn’t alone. Many of the issues raised by the Italian regulator are likely to cut to the core of all development of machine learning and generative AI systems, experts say. The EU is developing AI regulations, but so far there has been comparatively little action taken against the development of machine learning systems when it comes to privacy.

“There is this rot at the very foundations of the building blocks of this technology—and I think that’s going to be very hard to cure,” says Elizabeth Renieris, senior research associate at Oxford’s Institute for Ethics in AI and author on data practices. She points out that many data sets used for training machine learning systems have existed for years, and it is likely there were few privacy considerations when they were being put together. 

“There’s this layering and this complex supply chain of how that data ultimately makes its way into something like GPT-4,” Renieris says. “There’s never really been any type of data protection by design or default.” In 2022, the creators of one widely used image database, which has helped trained AI models for a decade, suggested images of people’s faces should be blurred in the data set. 

In Europe and California, privacy rules give people the ability to request that information be deleted or corrected if it is inaccurate. But deleting something from an AI system that is inaccurate or that someone doesn’t want there may not be straightforward—especially if the origins of the data are unclear. Both Renieris and Edwards question whether GDPR will be able to do anything about this in the long term, including upholding people’s rights. “There is no clue as to how you do that with these very large language models,” says Edwards from Newcastle University. “They don’t have provision for it.”

So far, there has been at least one relevant instance, when the company formerly known as Weight Watchers was ordered by the US Federal Trade Commission to delete algorithms created from data it didn’t have permission to use. But with increased scrutiny, such orders could become more common. “Depending, obviously, on the technical infrastructure, it may be difficult to fully clear your model of all of the personal data that was used to train it,” says Judin, from Norway’s data regulator. “If the model was then trained by unlawfully collected personal data, it would mean that you would essentially perhaps not be able to use your model.”

ChatGPT Has a Big Privacy Problem

Digging through manuals for security cameras, a group of gearheads found sinister details and ignited a new battle in the US-China tech war.

A Tiny Blog Took on Big Surveillance in China—and Won

North Korean hackers appear to have used the corrupted VoIP software to go after just a handful of crypto firms with “surgical precision.”

Software supply-chain attacks, in which hackers corrupt widely used applications to push their own code to thousands or even millions of machines, have become a scourge, both insidious and potentially huge in the breadth of their impact. But the latest major software supply-chain attack, in which hackers who appear to be working on behalf of the North Korean government hid their code in the installer for a common VoIP application known as 3CX, seems so far to have had a prosaic goal: breaking into a handful of cryptocurrency companies.

Researchers at Russian cybersecurity firm Kaspersky today revealed that they identified a small number of cryptocurrency-focused firms as at least some of the victims of the 3CX software supply-chain attack that's unfolded over the past week. Kaspersky declined to name any of those victim companies, but it notes that they're based in “western Asia.” 

Security firms CrowdStrike and SentinelOne last week pinned the operation on North Korean hackers, who compromised 3CX installer software that's used by 600,000 organizations worldwide, according to the vendor. Despite the potentially massive breadth of that attack, which SentinelOne dubbed “Smooth Operator,” Kaspersky has now found that the hackers combed through the victims infected with its corrupted software to ultimately target fewer than 10 machines—at least as far as Kaspersky could observe so far—and that they seemed to be focusing on cryptocurrency firms with “surgical precision.”

“This was all just to compromise a small group of companies, maybe not just in cryptocurrency, but what we see is that one of the interests of the attackers is cryptocurrency companies,” says Georgy Kucherin, a researcher on Kaspersky's GReAT team of security analysts. “Cryptocurrency companies should be especially concerned about this attack because they are the likely targets, and they should scan their systems for further compromise.”

Kaspersky based that conclusion on the discovery that, in some cases, the 3CX supply-chain hackers used their attack to ultimately plant a versatile backdoor program known as Gopuram on victim machines, which the researchers describe as “the final payload in the attack chain.” Kaspersky says the appearance of that malware also represents a North Korean fingerprint: It has seen Gopuram used before on the same network as another piece of malware, known as AppleJeus, linked to North Korean hackers. It's also previously seen Gopuram connect to the same command-and-control infrastructure as AppleJeus, and has seen Gopuram used previously to target cryptocurrency firms. All of that suggests not only that the 3CX attack was carried out by North Korean hackers, but that it may have been intended to breach cryptocurrency firms in order to steal from those companies, a common tactic of North Korean hackers ordered to raise money for the regime of Kim Jong-Un.

It has become a recurring theme for sophisticated state-sponsored hackers to exploit software supply chains to access the networks of thousands of organizations, only to winnow their focus down to a few victims. In 2020's notorious Solar Winds spy campaign, for instance, Russian hackers compromised the IT monitoring software Orion to push malicious updates to about 18,000 victims, but they appear to have stolen data from only a few dozen of them. In the earlier supply chain compromise of the CCleaner software, the Chinese hacker group known as Barium or WickedPanda compromised as many as 700,000 PCs, but similarly chose to target a relatively short list of tech firms.

“This is becoming very common,” says Kucherin, who also worked on the SolarWinds analysis and found clues linking that supply-chain attack to a known Russian group. “During supply-chain attacks, the threat actor conducts reconnaissance on the victims, collecting information, then they filter out this information, selecting victims to deploy a second-stage malware.” That filtering process is designed to help the hackers avoid detection, Kucherin points out, since deploying the second-stage malware to too many victims allows the attack to be more easily detected.

But Kucherin notes that the 3CX supply-chain attack was nonetheless detected relatively quickly, compared to others. The installation of the initial malware that the hackers appeared to use for reconnaissance was detected by companies like CrowdStrike and SentinelOne last week, less than a month after it was deployed. “They tried to be stealthy, but they failed,” Kucherin says. “Their first-stage implants were discovered.”

Given that detection, it's not clear how successful the campaign has been. Kucherin says Kaspersky hasn't seen any evidence of actual theft of cryptocurrency from the companies it saw targeted with the Gopuram malware.

But given the hundreds of thousands of potential victims of the 3CX supply-chain compromise, no one should conclude yet that crypto companies alone were targeted, says Tom Hegel, a security researcher with SentinelOne. “The current theory at this point is that the attackers did initially target crypto firms to get into those high-value organizations,” Hegel says. “I’m going to guess that once they saw the success of this, and the kinds of networks they were in, other objectives probably came into play.”

For now, Hegel says, no single security firm can see the whole shape of the 3CX hacking campaign or definitively state its goals. But if North Korean hackers really did compromise a piece of software used by 600,000 organizations around the world and use it just to try to steal cryptocurrency from a handful of them, they may have thrown away the keys to a much larger kingdom.

“This is all unfolding very quickly. I think we’ll continue to gain better insight into the victims,” Hegel says. “But from an attacker standpoint, if all they did was target crypto firms, this was a dramatic wasted opportunity.”

Massive 3CX Supply-Chain Hack Targeted Cryptocurrency Firms

An agency database WIRED obtained reveals widespread use of so-called 1509 summonses that experts say raises the specter of potential abuse.

US Immigration and Customs Enforcement agents are using an obscure legal tool to demand data from elementary schools, news organizations, and abortion clinics in ways that, some experts say, may be illegal. 

While these administrative subpoenas, known as 1509 custom summonses, are meant to be used only in criminal investigations about illegal imports or unpaid customs duties, WIRED found that the agency has deployed them to seek records that seemingly have little or nothing to do with customs violations, according to legal experts and several recipients of the 1509 summonses. 

A WIRED analysis of an Immigration and Customs Enforcement (ICE) subpoena tracking database, obtained through a Freedom of Information Act request, found that agents issued custom summons more than 170,000 times from the beginning of 2016 through mid-August 2022. The primary recipients of 1509s include telecommunications companies, major tech firms, money transfer services, airlines, and even utility companies. But it’s the edge cases that have drawn the most concern among legal experts, 

The outlier cases include custom summonses that sought records from a youth soccer league in Texas; surveillance video from a major abortion provider in Illinois; student records from an elementary school in Georgia; health records from a major state university’s student health services; data from three boards of elections or election departments; and data from a Lutheran organization that provides refugees with humanitarian and housing support.

In at least two instances, agents at ICE used the custom summons to pressure news organizations to reveal information about their sources. 

All of this is done without judicial oversight. 

“This is really dangerous,” says Matthew Guariglia, a policy analyst at the Electronic Frontier Foundation, a digital rights nonprofit. He adds that the frequent and apparently widespread use of 1509 summonses creates “a situation where this agency can fully go rogue” by using this tool in investigations that fall outside the scope of the law.

Two current agents from ICE’s Homeland Security Investigations (HSI) arm cautioned against drawing conclusions about potential abuse of the summons without additional context about each investigation referenced in the database. The agents, who asked not to be named because they are unauthorized to speak to the media, argue that because the agency is responsible for enforcing hundreds of laws pertaining to customs—including the distribution of child sexual abuse material (CSAM)—it’s possible that each summons was issued for a permissible investigation under the law. 

However, a former high-level Department of Homeland Security (DHS) official, who spoke on the condition of anonymity because their current employer forbids them from speaking to the press, expressed serious concern about many of the summonses. “It seems reasonable that ICE should be able to inspect records from a company like Amazon for customs investigations,” they say. “But what possible use or authority would ICE have for subpoenaing records from an abortion clinic?”

“It's like all power,” the official adds. “If it's not controlled, it gets abused.”

Casting a Wide Net

The 1509 customs summons is an administrative subpoena explicitly and exclusively meant for use in investigations of illegal imports or unpaid customs duties under a law known as Title 19 US Code 1509. Its goal is to provide agencies like ICE with a way to obtain business records from companies without having to go to a judge for a warrant. 

The subpoena-tracking database that WIRED obtained offers the most detailed breakdown of how ICE has been using the customs summons to date. The data shows that between January 4, 2016, and August 22, 2022, ICE issued 172,679 summonses, averaging more than 70 per day. Half of these were sent to telecommunications companies like AT&T, T-Mobile, and Comcast. Big technology companies like Google, Meta, and Microsoft collectively received nearly 15,000 summonses. Of the thousands of summonses sent to social media companies, Meta and Snapchat make up the vast majority.

Source

Wired