Apple has issued patches for several of its operating systems. The ones for iOS and iPadOS deserve your immediate attention.

Apple has released security patches for most of its operating systems, including iOS, Mac, iPadOS and watchOS.

Especially important are the updates for iOS and iPadOS which tackle vulnerabilities which could potentially leak sensitive user information. You should make sure you update as soon as you can.

To check if you’re using the latest software version, go to **Settings > General > Software Update**. It’s also worth turning on Automatic Updates if you haven’t already, which you can do on the same screen.

Update options

**Technical details**

Noteworthy are four vulnerabilities in Siri and another vulnerability in Accessibility which would allow an attacker with physical access to view sensitive user information. This may not seem very urgent at first, but if your device gets stolen then the thief can learn things about you which is far from ideal.

These are some of the vulnerabilities that jumped out at us.

CVE-2024-44274: a vulnerability in Accessibility that could allow an attacker with physical access to a locked device to view sensitive user information. This issue is fixed in iOS 17.7.1 and iPadOS 17.7.1, watchOS 11.1, iOS 18.1 and iPadOS 18.1 with improved authentication.

CVE-2024-44282: a vulnerability in Foundation where parsing a file could lead to disclosure of user information. This issue is fixed in tvOS 18.1, iOS 18.1 and iPadOS 18.1, iOS 17.7.1 and iPadOS 17.7.1, macOS Ventura 13.7.1, macOS Sonoma 14.7.1, watchOS 11.1, visionOS 2.1 by improved input validation. Foundation serves as a fundamental framework that offers a base layer of functionality for Apple’s operating systems. Among others it’s responsible for file system access.

CVE-2024-40867: a vulnerability in iTunes caused by a custom URL scheme handling issue that could be used by an attacker to break out of Web Content sandbox. This issue is fixed in iOS 18.1 and iPadOS 18.1 by improved input validation. Breaking out of the Web Content sandbox allows a malicious website or attacker to potentially access sensitive data, control other parts of the system, and compromise the overall security of the device beyond the intended limitations of the web browser.

**We don’t just report on phone security—we provide it**

Cybersecurity risks should never spread beyond a headline. Keep threats off your mobile devices by downloading Malwarebytes for iOS, and Malwarebytes for Android today.

 Update your iPhone, Mac, Watch: Apple issues patches for several vulnerabilities 

Find the best VPN for streaming with essential features like high-speed servers, strong encryption, streaming optimization, and broad…

Find the best VPN for streaming with essential features like high-speed servers, strong encryption, streaming optimization, and broad device compatibility. Enjoy seamless access to geo-restricted content securely.

Some geo-restricted content can only be unlocked and streamed using a Virtual Private Network (VPN). A VPN for streaming services allows you to watch content seamlessly while protecting your online identity.

With so many VPNs available, not all perform equally, making it challenging to find the best VPNs without interruptions. In this article, we’ll break down essential features to consider when selecting the top VPNs for streaming.

**Server Locations and Coverage**

An important feature in a streaming VPN is server locations and coverage, which determine the content you can access. Choose a VPN with a wide range of high-speed servers to prevent connectivity issues. A strong network spread across multiple countries will allow access to geo-restricted content. Look for VPN providers that frequently update servers, as this indicates a commitment to maintaining quality.

**Speed and Bandwidth**

To support HD streaming, ensure your VPN offers speeds of at least 100 Mbps, with no bandwidth limits or throttling. A VPN optimized for streaming should minimize buffering and reduce latency for smoother content loading.

****Encryption and Security****

When choosing a VPN for streaming, prioritize strong encryption and security. Look for AES-256 encryption or higher, and support for OpenVPN, IKEv2, and WireGuard protocols. The VPN should also include DNS leak protection to keep your IP address secure. If the VPN connection drops, an automatic internet disconnect feature ensures data remains private. These protections ensure that your data will not be exposed to third parties.

****Streaming Optimization****

Streaming optimization is key. The VPN should be optimized for high-speed, low-latency streaming, with custom protocols tailored to this purpose. Look for features that automatically select the best server for streaming, reducing buffering and enhancing speed.

****Device Compatibility****

Your VPN should support all the devices you plan to connect, including Windows, macOS, iOS, Android, and Linux. Look for compatibility with smart TVs (such as Samsung TV, Android TV, and Apple TV) and gaming consoles (PlayStation, Xbox, Nintendo Switch). It should also work with popular streaming platforms like Roku, Chromecast, and Amazon Fire TV. Browser extensions for Chrome, Firefox, and Safari add versatility, making it easier to stream across multiple devices.

****Simultaneous Connections****

Choose a VPN that supports at least five simultaneous connections. This feature lets you connect multiple devices without altering individual settings, which is convenient for families or business use. Check the VPN documentation for setup guides and confirm compatibility with less common devices like Linux and Chromebook. Additionally, look for split tunneling to configure specific device connections.

****Customer Support****

Reliable customer support is essential when using a VPN for streaming. Quick access to assistance for any issues or questions is critical. Look for providers that offer guides, FAQs, and tutorials for self-help, and a prompt email response time (ideally within 3 hours). Social media support is a plus. Effective customer support indicates a provider’s commitment to user satisfaction and technical support.

****Pricing and Payment Options****

Consider affordability and payment flexibility. Some VPNs offer discounts for long-term plans, free trials, money-back guarantees, and even cryptocurrency options. Make sure you’re getting value for your investment and review refund policies carefully. Look out for hidden fees or sudden price hikes, and ensure pricing transparency before subscribing.

****Logging Policy****

Opt for a VPN with a strict no-logs policy. It should avoid storing IP addresses, tracking online activities, or saving connection metadata. This policy helps ensure your privacy, reduces data breach risks, and underscores the VPN’s commitment to protecting client data.

****Compatibility with Popular Streaming Services****

Ensure the VPN is compatible with major streaming platforms you want to access, such as Netflix, Disney+, Hulu, Amazon Prime Video, HBO Max, and BBC iPlayer. Compatibility with these services ensures a smooth streaming experience and value for your subscription.

****Conclusion****

When selecting a VPN for streaming, consider these key features to make an informed choice. Look for speed, encryption, extensive server locations, streaming optimization, and privacy protections. Ensure the VPN provider doesn’t store user data and that its pricing, support, and device compatibility meet your expectations. Taking these steps will give you the best streaming experience while maintaining your online privacy.

1.  **Everything you need to know about VPN tracking**
2.  **Almost Every Major Free VPN Service is a Glorified Data Farm**
3.  **Surfshark VPN Data Breach Awareness with See-Through Toilet**
4.  **ASUS and NordVPN Partner to Integrate VPN Service into Routers**
5.  **Google Launches Verification Badges for Security Tested VPN Apps**

Top VPN Features to Consider When Choosing the Right Streaming Service

Apple unveils ‘Apple Intelligence’ for iPhone, iPad, and Mac devices while offering a $1 million bug bounty for…

Apple unveils ‘Apple Intelligence’ for iPhone, iPad, and Mac devices while offering a $1 million bug bounty for cybersecurity experts to find flaws in its Private Cloud Compute servers, enhancing privacy and AI security on iOS and macOS.

Apple today unveiled its highly anticipated AI-powered service, Apple Intelligence for iPhone, iPad, and Mac. This service, set to arrive with iOS 18.1, iPadOS 18.1, and macOS Sequoia 15.1, combines on-device processing with powerful cloud-based computing to handle complex AI tasks.

A preview of writing tools powered by Apple Intelligence

To ensure the security of these cloud-based systems, Apple has introduced a comprehensive bug bounty program. The tech giant is offering a reward of up to $1 million to security researchers who can successfully identify and exploit vulnerabilities in its “Private Cloud Compute” servers.

These servers, specifically designed for Apple Intelligence, are responsible for processing intensive AI tasks that cannot be handled by individual devices. This initiative is designed to strengthen the security of the Private Cloud Compute (PCC) servers that will underpin this service. The PCC servers, which will handle certain AI-powered tasks, are a critical component of Apple Intelligence. 

“We believe Private Cloud Compute is the most advanced security architecture ever deployed for cloud AI compute at scale, and we look forward to working with the research community to build trust in the system and make it even more secure and private over time,” Apple stated in its official statement.

Apple has taken significant steps to prioritize user privacy and security. To further strengthen trust, Apple has also opened up a Virtual Research Environment (VRE) to researchers and security enthusiasts.

The VRE provides access to the PCC software, allowing individuals to scrutinize its code and identify potential vulnerabilities. This allows them to scrutinize the system’s security measures and identify potential weaknesses.

The iPhone maker has also published a comprehensive security guide, detailing the intricacies of PCC’s security architecture and how it protects user data. This guide, coupled with the VRE, provides a robust platform for security researchers to delve deep into the system and uncover potential weaknesses.

The program offers a range of rewards, including $50,000 for accidental or unexpected data disclosure, $100,000 for uncertified code execution, $150,000 for access to user request data or sensitive information outside the trust boundary, and $250,000.

Additionally, $1,000 for arbitrary code execution without user permission or knowledge. Apple also promises to consider awarding money for any security issue that significantly impacts PCC, even if it doesn’t match a published category. 

By offering a substantial reward for critical vulnerabilities, Apple is encouraging the global cybersecurity community to contribute to the security of its AI infrastructure. The company is particularly interested in vulnerabilities that could compromise user data, execute code without permission, or exploit system design flaws.

By inviting the security community to scrutinize its infrastructure, Apple is also showing a commitment to transparency and security. To learn more about Apple Bug Bounty Program visit this link.

1.  Google Launches $250,000 kvmCTF Bug Bounty Program
2.  OpenAI Launches $200 to $20k ChatGPT Bug Bounty Program
3.  When Ethical Hackers Saved Companies from Devastating Hacks
4.  Multichain hack: Hacker returns $1m, keeps $150k as bug bounty
5.  A 19-year-old ethical hacker is a millionaire now; thanks to his skills

Apple Launches ‘Apple Intelligence’ and Offers $1M Bug Bounty for Security

LLMs tend to miss the forest for the trees, understanding specific instructions but not their broader context. Bad actors can take advantage of this myopia to get them to do malicious things, with a new prompt-injection technique.

Source: Igor Stevanovic via Alamy Stock Photo

A new prompt-injection technique could allow anyone to bypass the safety guardrails in OpenAI's most advanced language learning model (LLM).

GPT-4o, released May 13, is faster, more efficient, and more multifunctional than any of the previous models underpinning ChatGPT. It can process multiple different forms of input data in dozens of languages, then spit out a response in milliseconds. It can engage in real-time conversations, analyze live camera feeds, and maintain an understanding of context over extended conversations with users. When it comes to user-generated content management, however, GPT-4o is in some ways still archaic.

Marco Figueroa, generative AI (GenAI) bug-bounty programs manager at Mozilla, demonstrated in a new report how bad actors can leverage the power of GPT-4o while skipping over its guardrails. The key is to essentially distract the model by encoding malicious instructions in an unorthodox format, and spread them out in distinct steps.

**Tricking ChatGPT Into Writing Exploit Code**

To prevent malicious abuse, GPT-4o analyzes user inputs for any signs of bad language, instructions with ill intent, etc.

But at the end of the day, Figueroa says, "It's just word filters. That's what I've seen through experience, and we know exactly how to bypass these filters."

For example, he says, "We can modify how something's spelled out — break it up in certain ways — and the LLM interprets it." GPT-4o might not reject a malicious instruction if it's presented with a spelling or phrasing that doesn't accord with typical natural language.

Figuring out the exact right way to present information in order to dupe state-of-the-art AI, though, requires lots of creative brain power. It turns out that there's a much simpler method for bypassing GPT-4o's content filtering: encoding instructions in a format other than natural language.

To demonstrate, Figueroa arranged an experiment with the goal of getting ChatGPT to do something it otherwise shouldn't: write exploit code for a software vulnerability. He picked CVE-2024-41110, a bypass for authorization plug-ins in Docker that earned a "critical" 9.9 out of 10 rating in the Common Vulnerability Scoring System (CVSS) this summer.

To trick the model, he encoded his malicious input in hexadecimal format, and provided a set of instructions for decoding it. GPT-4o took that input — a long series of digits and letters A through F — and followed those instructions, ultimately decoding the message as an instruction to research CVE-2024-41110 and write a Python exploit for it. To make it less likely that the program would make a fuss over that instruction, he used some leet speak, asking for an "3xploit," instead of an "exploit."

Source: Mozilla

In a minute flat, ChatGPT generated a working exploit similar to, but not exactly like, another PoC already published to GitHub. Then, as a bonus, it attempted to execute the code against itself. "There wasn't any instruction that specifically said to execute it. I just wanted to print it out. I didn't even know why it went ahead and did that," Figueroa says.

**What's Missing in GPT-4o?**

It's not just that GPT-4o is getting distracted by decoding, according to Figueroa, but that it's in some sense missing the forest for the trees — a phenomenon that has been documented in other prompt-injection techniques lately.

"The language model is designed to follow instructions step-by-step, but lacks deep context awareness to evaluate the safety of each individual step in the broader context of its ultimate goal," he wrote in the report. The model analyzes each input — which, on its own, doesn't immediately read as harmful — but not what the inputs produce in sum. Rather than stop and think about how instruction one bears on instruction two, it just charges ahead.

"This compartmentalized execution of tasks allows attackers to exploit the model's efficiency at following instructions without deeper analysis of the overall outcome," according to Figueroa.

If this is the case, ChatGPT will not only need to improve how it handles encoded information but also develop a kind of broader context around instructions split into distinct steps.

To Figueroa, though, OpenAI appears to have been valuing innovation at the cost of security when developing its programs. "To me, they don't care. It just feels like that," he says. By contrast, he's had much more trouble trying the same jailbreaking tactics against models by Anthropic, another prominent AI company founded by former OpenAI employees. "Anthropic has the strongest security because they have built both a prompt firewall \[for analyzing inputs\] and response filter \[for analyzing outputs\], so this becomes 10 times more difficult," he explains.

Dark Reading is awaiting comment from OpenAI on this story.

**About the Author**

Nate Nelson is a freelance writer based in New York City. Formerly a reporter at Threatpost, he contributes to a number of cybersecurity blogs and podcasts. He writes "Malicious Life" -- an award-winning Top 20 tech podcast on Apple and Spotify -- and hosts every other episode, featuring interviews with leading voices in security. He also co-hosts "The Industrial Security Podcast," the most popular show in its field.

Mozilla: ChatGPT Can Be Manipulated Using Hex Code

Plus: Apple offers $1 million to hack its AI cloud infrastructure, Iranian hackers successfully peddle stolen Trump campaign docs, Russia hacks the nation of Georgia, and a “cyberattack” that wasn’t.

The Chinese spy operation adds to the growing sense of a melee of foreign digital interference in the election, which has already included Iranian hackers’ attempt to hack and leak emails from the Trump campaign—with limited success—and Russia-linked disinformation efforts across social media.

**Apple Releases Security Research Tools for Private Cloud Compute**

Ahead of the full launch next week of Apple’s AI platform, Apple Intelligence, the company debuted tools this week for security researchers to evaluate its cloud infrastructure known as Private Cloud Compute. Apple has gone to great lengths to engineer a secure and private AI cloud platform, and this week’s release includes extensive detailed technical documentation of its security features as well as a research environment that is already available in the macOS Sequoia 15.1 beta release. The testing features allow researchers (or anyone) to download and evaluate the actual version of PCC software that Apple is running in the cloud at a given time. The company tells WIRED that the only modifications to the software relate to optimizing it to run in the virtual machine for the research environment. Apple also released the PCC source code and said that as part of its bug bounty program, vulnerabilities that researchers discover in PCC will be eligible for a maximum bounty payout of up to $1 million.

**Iranian Hackers Found Takers for Their Stolen Trump Emails**

Over the summer, Politico, The New York Times, and The Washington Post each revealed that they’d been approached by a source offering hacked Trump campaign emails—a source whom the US Justice Department says was working on behalf of the Iranian government. The news outlets all refused to publish or report on those stolen materials. Now it appears that Iran’s hackers did eventually find outlets outside the mainstream media that were willing to release those emails. American Muckrakers, a PAC run by a Democratic operative, did publish the documents after soliciting them in a public post on X, writing, “Send it to us and we'll get it out.”

American Muckrakers then published internal Trump campaign communications about North Carolina Republican gubernatorial candidate Mark Robinson and Florida Republican representative Anna Paulina Luna, as well as material that seemed to suggest a financial arrangement between Donald Trump and Robert F. Kennedy Jr., the third-party candidate who dropped out of the race and endorsed Trump. Independent journalist Ken Klippenstein also received and published some of the hacked material, including a research profile on Trump running mate and US senator J.D. Vance that the campaign assembled when assessing him for the role. Klippenstein subsequently received a visit from the FBI, he’s said, warning him that the documents were shared as part of a foreign influence campaign. Klippenstein has defended his position, arguing that the media should not serve as "gatekeeper of what the public should know.”

**Russian Cyberspies Hacked the Entire Nation of Georgia**

As Russia has both waged war and cyberwar against Ukraine, it’s also carried out a vast campaign of hacking against another neighbor to the West with whom it’s long had a fraught relationship: Georgia. Bloomberg this week revealed ahead of the Georgian election how Russia systematically penetrated the smaller country’s infrastructure and government in a yearslong series of digital intrusion operations. From 2017 to 2020, for instance, Russia’s military intelligence agency, the GRU, hacked Georgia’s Central Election Commission (just as it did in Ukraine in 2014), multiple media organizations, and IT systems at the country’s national railway company—all in addition to the attack on Georgian TV stations that the NSA pinned on the GRU’s Sandworm unit in 2020. Meanwhile, hackers known as Turla, working for the Kremlin’s KGB successor the FSB, broke into Georgia’s Foreign Ministry and stole gigabytes of officials’ emails over months. According to Bloomberg, Russia’s hacking efforts weren’t limited to espionage, but also appeared to include preparing for disruption of Georgian infrastructure like the electric grid and oil companies in the event of an escalating conflict.

**This May Be the Worst-Ever Headline About a “Cyberattack”**

For years, cybersecurity professionals have argued about what constitutes a cyberattack. An intrusion designed to destroy data, cause disruption, or sabotage infrastructure? Yes, that’s a cyberattack. A hacker breach to steal data? No. A hack-and-leak operation or an espionage mission with a disruptive clean-up phase? Probably not, but there’s room for debate. The Jerusalem Post this week, however, achieved perhaps the clearest-cut example of calling something a cyberattack—in a headline no less—that is very clearly not: disinformation on social media. The so-called “Hezbollah cyberattack” that the news outlet reported was a collection of photos of Israeli hospitals posted by “hackers” identifying as Hezbollah supporters that suggested weapons and cash were stored underneath them and that they should be attacked. The posts seemingly came in response to the Israeli Defense Forces’ repeating similar claims about hospitals in Gaza that the IDF has bombed, as well as another more recently in Lebanon’s capital city of Beirut.

“These are NOT CYBERATTACKS,” security researcher Lukasz Olejnik, the author of the books _The Philosophy of Cybersecurity_ and _Propaganda_, wrote next to a screenshot of the Jerusalem Post headline on X. “Posting images to social media is not hacking. Such a bad take.”

Chinese Hackers Target Trump Campaign via Verizon Breach

Kremlin intelligence carried out a wide-scale phishing campaign in contrast to its usual, more targeted operations.

Source: Design Pics Inc via Alamy Stock Photo

Russia's premiere advanced persistent threat group has been phishing thousands of targets in militaries, public authorities, and enterprises.

APT29 (aka Midnight Blizzard, Nobelium, Cozy Bear) is arguably the world's most notorious threat actor. An arm of the Russian Federation's Foreign Intelligence Service (SVR), it's best known for the historic breaches of SolarWinds and the Democratic National Committee (DNC). Lately, it has breached Microsoft's codebase and political targets across Europe, Africa, and beyond.

"APT29 embodies the 'persistent' part of 'advanced persistent threat,'" says Satnam Narang, senior staff research engineer at Tenable. "It has persistently targeted organizations in the United States and Europe for years, utilizing various techniques, including spear-phishing and exploitation of vulnerabilities to gain initial access and elevate privileges. Its modus operandi is the collection of foreign intelligence, as well as maintaining persistence in compromised organizations in order to conduct future operations."

Along these same lines, the Computer Emergency Response Team of Ukraine (CERT-UA) recently discovered APT29 phishing Windows credentials from government, military, and private sector targets in Ukraine. And after comparing notes with authorities in other countries, CERT-UA found that the campaign was actually spread across "a wide geography."

That APT29 would go after sensitive credentials from geopolitically prominent and diverse organizations is no surprise, Narang notes, though he adds that "the one thing that does kind of stray from the path would be its broad targeting, versus \[its typical more\] narrowly focused attacks."

**AWS and Microsoft**

The campaign, which dates back to August, was carried out using malicious domain names designed to seem like they were associated with Amazon Web Services (AWS). The emails sent from these domains pretended to advise recipients on how to integrate AWS with Microsoft services, and how to implement zero trust architecture.

Despite the masquerade, AWS itself reported that the attackers weren't after Amazon, or its customers' AWS credentials.

What APT29 really wanted was revealed in the attachments to those emails: configuration files for Remote Desktop, Microsoft's application for implementing the Remote Desktop Protocol (RDP). RDP is a popular tool that legitimate users and hackers alike use to operate computers remotely.

"Normally, attackers will try to brute force their way into your system or exploit vulnerabilities, then have RDP configured. In this case, they're basically saying: 'We want to establish that connection \[upfront\],'" Narang says.

Launching one of these malicious attachments would have immediately triggered an outgoing RDP connection to an APT29 server. But that wasn't all: The files also contained a number of other malicious parameters, such that when a connection was made, the attacker was given access to the target computer's storage, clipboard, audio devices, network resources, printers, communication (COM) ports, and more, with the added ability to run custom malicious scripts.

**Block RDP**

APT29 may not have used any legitimate AWS domains, but Amazon still managed to interrupt the campaign by seizing the group's malicious copycats.

For potential victims, CERT-UA recommends strict precautions: not just monitoring network logs for connections to IP addresses tied to APT29 but also analyzing all outgoing connections to all IP addresses on the wider Web through the end of the month.

And for organizations at risk in the future, Narang offers simpler advice. "First and foremost, don't allow RDP files to be received. You can block them at your email gateway. That's going to kneecap this whole thing," he says.

AWS declined to provide further comment for this story. Dark Reading has also reached out to Microsoft for its perspective.

**About the Author**

Nate Nelson is a freelance writer based in New York City. Formerly a reporter at Threatpost, he contributes to a number of cybersecurity blogs and podcasts. He writes "Malicious Life" -- an award-winning Top 20 tech podcast on Apple and Spotify -- and hosts every other episode, featuring interviews with leading voices in security. He also co-hosts "The Industrial Security Podcast," the most popular show in its field.

Russia's APT29 Mimics AWS Domains to Steal Windows Credentials

Apple has publicly made available its Private Cloud Compute (PCC) Virtual Research Environment (VRE), allowing the research community to inspect and verify the privacy and security guarantees of its offering.
PCC, which Apple unveiled earlier this June, has been marketed as the "most advanced security architecture ever deployed for cloud AI compute at scale." With the new technology, the idea is

Cloud Security / Artificial Intelligence

Apple has publicly made available its Private Cloud Compute (PCC) Virtual Research Environment (VRE), allowing the research community to inspect and verify the privacy and security guarantees of its offering.

PCC, which Apple unveiled earlier this June, has been marketed as the "most advanced security architecture ever deployed for cloud AI compute at scale." With the new technology, the idea is to offload computationally complex Apple Intelligence requests to the cloud in a manner that doesn't sacrifice user privacy.

Apple said it's inviting "all security and privacy researchers — or anyone with interest and a technical curiosity — to learn more about PCC and perform their own independent verification of our claims."

To further incentivize research, the iPhone maker said it's expanding the Apple Security Bounty program to include PCC by offering monetary payouts ranging from $50,000 to $1,000,000 for security vulnerabilities identified in it.

This includes flaws that could allow execution of malicious code on the server, and exploits capable of extracting users' sensitive data, or information about the user's requests.

The VRE aims to offer a suite of tools to help researchers carry out their analysis of PCC from the Mac. It comes with a virtual Secure Enclave Processor (SEP) and leverages built-in macOS support for paravirtualized graphics to enable inference.

Apple also said it's making the source code associated with some components of PCC accessible via GitHub to facilitate a deeper analysis. This includes CloudAttestation, Thimble, splunkloggingd, and srd\_tools.

"We designed Private Cloud Compute as part of Apple Intelligence to take an extraordinary step forward for privacy in AI," the Cupertino-based company said. "This includes providing verifiable transparency – a unique property that sets it apart from other server-based AI approaches."

The development comes as broader research into generative artificial intelligence (AI) continues to uncover novel ways to jailbreak large language models (LLMs) and produce unintended output.

Earlier this week, Palo Alto Networks detailed a technique called Deceptive Delight that involves mixing malicious and benign queries together to trick AI chatbots into bypassing their guardrails by taking advantage of their limited "attention span."

The attack requires a minimum of two interactions, and works by first asking the chatbot to logically connect several events – including a restricted topic (e.g., how to make a bomb) – and then asking it to elaborate on the details of each event.

Researchers have also demonstrated what's called a ConfusedPilot attack, which targets Retrieval-Augmented Generation (RAG) based AI systems like Microsoft 365 Copilot by poisoning the data environment with a seemingly innocuous document containing specifically crafted strings.

"This attack allows manipulation of AI responses simply by adding malicious content to any documents the AI system might reference, potentially leading to widespread misinformation and compromised decision-making processes within the organization," Symmetry Systems said.

Separately, it has been found that it's possible to tamper with a machine learning model's computational graph to plant "codeless, surreptitious" backdoors in pre-trained models like ResNet, YOLO, and Phi-3, a technique codenamed ShadowLogic.

"Backdoors created using this technique will persist through fine-tuning, meaning foundation models can be hijacked to trigger attacker-defined behavior in any downstream application when a trigger input is received, making this attack technique a high-impact AI supply chain risk," Hidden Layer researchers Eoin Wickens, Kasimir Schulz, and Tom Bonner said.

"Unlike standard software backdoors that rely on executing malicious code, these backdoors are embedded within the very structure of the model, making them more challenging to detect and mitigate."

Found this article interesting? Follow us on Twitter __ and LinkedIn to read more exclusive content we post.

Apple Opens PCC Source Code for Researchers to Identify Bugs in Cloud AI Security

An attacker compromised one of Fortinet's most sensitive products and mopped up all kinds of reconnaissance data helpful for future mass device attacks.

Source: Thomas Kyhn via Alamy Stock Photo

An unknown threat actor has compromised Fortinet devices en masse across various industries, leaving no particular indication of what they plan to do next.

The campaign was enabled by a critical vulnerability, CVE-2024-47575, which the Cybersecurity and Infrastructure Security Agency (CISA) has just added to its Known Exploited Vulnerability (KEV) catalog. It affects Fortinet's FortiManager tool, the single, centralized console from which organizations can manage all their Fortinet brand firewalls, access points, application delivery controllers (ADCs), and email gateways. Up to 100,000 devices can be managed from a single FortiManager interface, making it an efficient tool for IT administration, and a spectacular launchpoint for cyberattacks.

According to Mandiant, a threat actor it now tracks as UNC5820 used CVE-2024-47575 to compromise more than 50 instances of FortiManager. Doing so enabled them to siphon off information about the various devices connected to those FortiManager instances, which could prove useful in follow-on attacks. To this point, however, no malicious follow-on activity has been observed.

**A Critical Vulnerability in FortiManager**

CVE-2024-47575 results from a missing authentication in the fgfmd daemon, a "critical function" that facilitates communication between FortiManager and the various Fortinet devices it manages. Using specially crafted requests, a remote, unauthenticated attacker could exploit this missing authentication to execute arbitrary code or commands in a targeted device. The centrality of the vulnerable daemon, combined with the severe effect of such an attack, have earned CVE-2024-47575 a "critical" 9.8 out of 10 score according to the Common Vulnerability Scoring System (CVSS).

“These types of exploits are some of the most coveted by attackers as little to no action is required on the part of the victim for the attacker to gain remote access," notes T. Frank Downs, senior director of proactive services at BlueVoyant. "Large-scale exploitation could enable lateral movement to other managed devices, leading to widespread network disruption and data breaches. These actions, in turn, could allow attackers to exfiltrate sensitive data from FortiManager devices, including configurations and credentials."

The unidentified threat actor UNC5820 has already halfway demonstrated what one can do with CVE-2024-47575. Beginning June 27, UNC5820 connected to multiple Fortinet devices from an IP address in Japan. Quickly, a series of important files were zipped into an archive file. These included the targeted FortiManager's build, version, and branch data, configuration files for FortiGate devices it managed, hashed passwords, and more.

Researchers identified another exploitation attempt in September, during which the attacker managed to register their own, unauthorized Fortinet device to the targeted FortiManager console.

In theory, all this data would have been useful for getting to know the target's environment, and laying the groundwork for a mean follow-on attack. "The potential damage from this vulnerability is significant, as it allows attackers to remotely execute code and access sensitive data without authentication, leading to data breaches, network disruptions, and unauthorized access to critical systems," Downs says. And yet, Mandiant has not observed evidence of any such attacks to date.

**What to Do Now**

To exploit CVE-2024-47575 in the first place, UNC5820 would have required some means to reach an organization's FortiManager device. Thus, only those exposed to the Internet are likely to have been targeted.

For organizations with exposed management consoles, Mandiant recommends immediate, thorough forensic investigations, and only allowing known devices and IPs access to FortiManager. Fortinet's FortiGuard Labs has also published further recommendations for remediation to its blog, including workarounds for cases in which an upgrade to the latest software is not possible.

In response to a request for comment from Dark Reading, Fortinet offered the following statement:

"After identifying this vulnerability (CVE-2024-47575), Fortinet promptly communicated critical information and resources to customers. This is in line with our processes and best practices for responsible disclosure to enable customers to strengthen their security posture prior to an advisory being publicly released to a broader audience, including threat actors. We also have published a corresponding public advisory (FG-IR-24-423) reiterating mitigation guidance, including a workaround and patch updates. We urge customers to follow the guidance provided to implement the workarounds and fixes and to continue tracking our advisory page for updates. We continue to coordinate with the appropriate international government agencies and industry threat organizations as part of our ongoing response."

**About the Author**

Nate Nelson is a freelance writer based in New York City. Formerly a reporter at Threatpost, he contributes to a number of cybersecurity blogs and podcasts. He writes "Malicious Life" -- an award-winning Top 20 tech podcast on Apple and Spotify -- and hosts every other episode, featuring interviews with leading voices in security. He also co-hosts "The Industrial Security Podcast," the most popular show in its field.

Critical Bug Exploited in Fortinet's Management Console

The Russian-language malware primarily enlists computers to mine Monero, but theoretically it can do worse.

Source: Artimages via Alamy Stock Photo

An 8-year-old modular botnet is still kicking, spreading a cryptojacker and Web shell on machines spread across multiple continents.

"Prometei" was first discovered in 2020, but later evidence suggested that it's been in the wild since at least 2016. In those intervening years it spread to more than 10,000 computers globally, in countries as diverse as Brazil, Indonesia, Turkey, and Germany, whose Federal Office for Information Security categorizes it as a medium-impact threat.

"Prometei's reach is global due to its focus on widely used software vulnerabilities," explains Callie Guenther, senior manager of cyber-threat research at Critical Start. "The botnet spreads through weak configurations and unpatched systems, targeting regions with inadequate cybersecurity practices. Botnets like Prometei typically do not discriminate by region but seek maximum impact by exploiting systemic weaknesses. \[In this case\], organizations using unpatched or poorly configured Exchange servers are particularly at risk."

Trend Micro details what a Prometei attack looks like: clunky in its initial infection but stealthy thereafter, capable of exploiting vulnerabilities in a variety of different services and systems, and focused on cryptojacking but capable of more.

**Loud Entry Into Unloved Systems**

Don't expect an initial Prometei infection to be terribly sophisticated.

The case Trend Micro observed began with a number of failed network login attempts from two IP addresses appearing to come from Cape Town, South Africa, which aligned closely with known Prometei infrastructure.

After its first successful login into a machine, the malware went to work testing out a variety of outdated vulnerabilities that might still be lingering in its target's environment. For example, it uses the half-decade old "BlueKeep" bug in the Remote Desktop Protocol (RDP) — rated a "critical" 9.8 out of 10 in the Common Vulnerability Scoring System — to try and achieve remote code execution (RCE). It uses the even older EternalBlue vulnerability to propagate via Server Message Block (SMB). On Windows systems, it tries the 3-year-old ProxyLogon arbitrary file write vulnerabilities CVE-2021-27065 and CVE-2021-26858, which have "high" 7.8 CVSS ratings.

Exploiting such old vulnerabilities could be read as lazy. In another light, it's an effective approach to weeding out better-equipped systems belonging to more active organizations.

"Prime targets are those systems that have not been or cannot be patched for some reason, which translates to them being either unmonitored or neglected from normal security processes," Mayuresh Dani, manager of security research at Qualys, points out. "The malware authors want to go after easy pickings, and in today's connected world, I consider this intelligent, as if they know that their targets will be plagued by multiple security issues."

**Prometei's Fire**

Once Prometei gets to where it wants to go, it has some neat tricks for achieving its ends. It uses a domain generation algorithm (DGA) to harden its command-and-control (C2) infrastructure, enabling it to continue operating even if victims try blocking one or more of its domains. It manipulates targeted systems to allow its traffic through firewalls, and runs itself automatically upon system reboots.

One particularly useful Prometei command evokes the WDigest authentication protocol, which stores passwords in plaintext in memory. WDigest is typically disabled in modern Windows systems, so Prometei forces those plaintext passwords, which it then dumps into a dynamic link library (DLL). Then, another Prometei command configures Windows Defender to ignore that particular DLL, allowing those passwords to be exfiltrated without raising any red flags.

The most obvious purpose of a Prometei infection appears to be cryptojacking — using infected machines to help mine the ultra-anonymous Monero cryptocurrency without their owners' knowing it. Beyond that, though, it downloads and configures an Apache Web server that serves as a persistent Web shell. The Web shell allows attackers to upload more malicious files and execute arbitrary commands.

As Stephen Hilt, senior threat researcher at Trend Micro, points out, botnet infections are often associated with other kinds of attacks as well.

"I always look at the cryptomining groups being a canary in the coal mine — it's an indicator that there's probably more going on in your system," he says. "If you look at our 2021 blog, there was LemonDuck, a ransomware group, and \[Prometei\] all within the same machines."

**Russia Links**

There is one specific part of the globe that Prometei does not touch.

The botnet's Tor-based C2 server is made to specifically avoid certain exit nodes in some former Soviet countries. To further ensure the safety of Russian-language targets, it possesses a credential-stealing component that deliberately avoids affecting any accounts labeled "Guest" or "Other user" in Russian.

Older variants of the malware contained bits of Russian-language settings and language code, and the name "Prometei" is a translation of "Prometheus" in various Slavic languages. In the famous myth, Zeus programs an eagle to attack Prometheus' liver every day, only for the liver to persist through reboots each night.

**About the Author**

Nate Nelson is a freelance writer based in New York City. Formerly a reporter at Threatpost, he contributes to a number of cybersecurity blogs and podcasts. He writes "Malicious Life" -- an award-winning Top 20 tech podcast on Apple and Spotify -- and hosts every other episode, featuring interviews with leading voices in security. He also co-hosts "The Industrial Security Podcast," the most popular show in its field.

'Prometei' Botnet Spreads Its Cryptojacker Worldwide

Popular titles on both Google Play and Apple's App Store include hardcoded and unencrypted AWS and Azure credentials in their codebases or binaries, making them vulnerable to misuse by threat actors.

Source: Aleksia via Alamy Stock Photo

Several widely used mobile apps, some with millions of downloads, expose hardcoded and unencrypted credentials to cloud services within their code bases, researchers from Symantec have found. This potentially allows anyone with access to the app’s binary or source code to extract the credentials to exploit cloud infrastructure for misuse.

Popular apps for both Android and iPhone devices include credentials for either Amazon Web Services (AWS) and Microsoft Azure Blog Storage within their code, Symantec revealed in a blog post this week. And they're found on each device platform’s respective official mobile app store: Google Play and Apple's App Store.

"This dangerous practice means that anyone with access to the app's binary or source code could potentially extract these credentials and misuse them to manipulate or exfiltrate data, leading to severe security breaches," Symantec engineers wrote in the post.

Further, the "widespread nature" of the vulnerabilities across apps for both iOS and Android platforms "underscores the urgent need for a shift towards more secure development practices" when it comes to mobile applications, they added.

Symantec’s research zeroed in on a number of widely distributed mobile applications that included either AWS or Azure credentials in their codebases. In terms of the former, both Android and iOS apps are guilty of credential exposure, while several Android apps expose Azure storage credentials.

Related:SoftwareOne Launches Cloud Competency Centre in Malaysia

For example, an app called The Pic Stitch: Collage Maker found on the Google Play store contains hardcoded AWS production credentials — including the production Amazon S3 bucket name, the read and write access keys, and secret keys — in its codebase, the researchers found. It also reveals staging credentials in some cases.

**iOS Apps With Serious Security Risks**

Meanwhile, three iOS apps examined by Symantec also were found to expose AWS credentials. One called Crumbl, which has more than 3.9 million user ratings and is ranked No. 5 in the Food & Drink category on the Apple App Store, initializes an AWSStaticCredentialsProvider with plaintext credentials. The credentials, which are used to configure AWS services, include both an access key and secret key.

Furthermore, the app also includes another "significant security oversight" by including a WebSocket Secure (WSS) endpoint within its code. This endpoint, part of the Amazon API URL, is hardcoded with an API Gateway that directly connects to the Internet of Things services on AWS.

"Exposing such URLs alongside static credentials makes it easier for attackers to potentially intercept or manipulate communications, leading to unauthorized access to the associated AWS resources," the engineers wrote. Thus, this vulnerable configuration, without proper encryption or obfuscation, "presents a serious risk to the integrity of the application and its backend infrastructure," they noted.

Related:Unmanaged Cloud Credentials Pose Risk to Half of Orgs

Two other iOS apps with hundreds of thousands of App Store ratings also expose AWS credentials by hardcoding them directly within their code; the apps are Eureka: Earn Money for Surveys and Videoshop – Video Editor.

The former allocates an INMAWSCredentials object and initializes it with the access key and secret key, both stored in plaintext and which can be used to log events to AWS, "exposing critical cloud resources to potential attacks," the engineers said.

The latter directly embeds unencrypted AWS credentials in the \[VSAppDelegate setupS3\] method, which means anyone with access to the app's binary could easily extract them. This would give them unauthorized access to the associated S3 buckets and potentially lead to data theft or manipulation.

**Android Apps Expose Azure Credentials**

Similarly, three Android applications expose credentials to Microsoft Azure Blob Storage directly, via either their binaries or codebases, Symantec found.

Related:Cisco Disables DevHub Access After Security Breach

An Indian ride-sharing app, Meru Cabs — which has more than 5 million downloads on Google Play — includes hardcoded Azure credentials within its UploadLogs service by embedding a connection string that includes an account key. "This connection string is used to manage log uploads, exposing critical cloud storage resources to potential abuse," the engineers wrote.

Sulekha Business, another Android app with more than 500,000 downloads, embeds multiple hardcoded Azure credentials used for various purposes — such as adding posts, handling invoices, and storing user profiles — across its codebase.

A third Android app that also has more than 500,000 downloads, ReSound Tinnitus Relief, also hardcodes Azure Blob Storage credentials for managing various assets and sound files, the exposure of which could lead to unauthorized access and data breaches.

**Mitigation Begins With App Development**

Symantec’s findings come a day after the release of a report by Datadog that found that unmanaged credentials that live for too long on a cloud-based network posed a security risk to half of organizations. Indeed, any inadvertent disclosure of credentials to cloud services exposes any organization with network infrastructure, software, or other assets running on them to significant risk, according to Symantec.

A good place to start to mitigate these risks is in the development of applications, where developers should follow best practices for managing sensitive information. They include the use of environment variables to store sensitive credentials so they are loaded at runtime rather than embedded directly in the app's code, according to Symantec.

Developers also should use dedicated secrets management tools, such as AWS Secrets Manager or Azure Key Vault, to securely store and access credentials. If the credentials must be stored in the app, then they should ensure that they use strong encryption algorithms, and decrypt them at runtime as needed.

According to Symantec, another way to protect credentials and also avoid other potential app-development missteps is to integrate automated security-scanning tools into the development pipeline to detect common security flaws early in the development process.

**About the Author**

Elizabeth Montalbano is a freelance writer, journalist, and therapeutic writing mentor with more than 25 years of professional experience. Her areas of expertise include technology, business, and culture. Elizabeth previously lived and worked as a full-time journalist in Phoenix, San Francisco, and New York City; she currently resides in a village on the southwest coast of Portugal. In her free time, she enjoys surfing, hiking with her dogs, traveling, playing music, yoga, and cooking.

Tag

#apple