UNAPIMON works by meticulously disabling hooks in Windows APIs for detecting malicious processes.

Source: Panchenko Vladimir via Shutterstock

Researchers have spotted Earth Freybug, a China-linked threat actor, using a new malware tool to bypass mechanisms organizations might have put in place to monitor Windows application programming interfaces (APIs) for malicious activity.

The malware, which researchers at Trend Micro discovered and named UNAPIMON, works by disabling hooks in Windows APIs for inspecting and analyzing API-related processes for security issues.

**Unhooking APIs**

The goal is to prevent any processes that the malware spawns from being detected or inspected by antivirus tools, sandboxing products, and other threat detection mechanisms.

"Looking at the behavior of UNAPIMON and how it was used in the attack, we can infer that its primary purpose is to unhook critical API functions in any child process," Trend Micro said in a report this week.

"For environments that implement API monitoring through hooking, such as sandboxing systems, UNAPIMON will prevent child processes from being monitored," the security vendor said. This allows malicious programs to run without being detected.

Trend Micro assessed Earth Freybug as being a subset of APT41, a collective of Chinese threat groups variously referred to as Winnti, Wicked Panda, Barium, and Suckfly. The group is known for using a collection of custom tools and so-called living-off-the-land binaries (LOLbins) that manipulate legitimate system binaries such as PowerShell and Windows Management Instrumentation (WMI).

APT41 itself has been active since at least 2012 and is linked to numerous cyber espionage campaigns, supply chain attacks, and financially motivated cybercrime. In 2022, researchers at Cybereason identified the threat actor as stealing large volumes of trade secrets and intellectual property from companies in the US and Asia for years. Its victims have included manufacturing and IT organizations, governments, and critical infrastructure targets in the US, East Asia, and Europe. In 2020, the US government charged five members believed to be associated with the group for their role in attacks against more than 100 organizations globally.

**Attack Chain**

In the recent incident that Trend Micro observed, Earth Freybug actors used a multistaged approach to delivering UNAPIMON on target systems. In the first stage, the attackers injected malicious code of unknown origin into vmstools.exe, a process associated with a set of utilities for facilitating communications between a guest virtual machine and the underlying host machine. The malicious code created a scheduled task on the host machine to run a batch script file (cc.bat) on the host system.

The batch file's task is to collect a range of system information and initiate a second scheduled task to run a cc.bat file on the infected host. The second batch script file leverages SessionEnv, a Windows service for managing remote desktop services, to side-load a malicious dynamic link library (DLL) on the infected host. "The second cc.bat is notable for leveraging a service that loads a nonexistent library to side-load a malicious DLL. In this case, the service is SessionEnv," Trend Micro said.

The malicious DLL then drops UNAPIMON on the Windows service for defense evasion purposes and also on a cmd.exe process that quietly executes commands. "UNAPIMON itself is straightforward: It is a DLL malware written in C++ and is neither packed nor obfuscated; it is not encrypted save for a single string," Trend Micro said. What makes it "peculiar" is its defense evasion technique of unhooking APIs so that the malware's malicious processes remain invisible to threat detection tools. "In typical scenarios, it is the malware that does the hooking. However, it is the opposite in this case," Trend Micro said.

**About the Author(s)**

Jai Vijayan is a seasoned technology reporter with over 20 years of experience in IT trade journalism. He was most recently a Senior Editor at Computerworld, where he covered information security and data privacy issues for the publication. Over the course of his 20-year career at Computerworld, Jai also covered a variety of other technology topics, including big data, Hadoop, Internet of Things, e-voting, and data analytics. Prior to Computerworld, Jai covered technology issues for The Economic Times in Bangalore, India. Jai has a Master's degree in Statistics and lives in Naperville, Ill.

China-Linked Threat Actor Taps 'Peculiar' Malware to Evade Detection

Fortanix is working on technologies to build a security wall around AI search.

The inner functioning of AI-powered search is more complex than plain-text search via Google when it comes to extracting answers from websites and databases. Fortanix is trying to build a security wall to protect the search query and the extraction of data from artificial intelligence (AI) systems.

Generative AI technologies will ultimately provide fine-grained responses based on associativity of information and retrieve information from sources users may not be aware of, says Richard Searle, vice president of confidential computing at Fortanix.

"What we're finding ... is that there is a deeper focus happening in the AI domain around privacy, consent, and permissioning of information," Searle says. "These are obviously the core cases."

Essentially, Fortanix is building a security wall around the way AI search queries are handled. More specifically, it is building security around prompts, where users provide AI search queries. The security wall extends to data retrieval from large language models, which are delivered to customers.

"We think there's a significant market there," Searle says. "The partners that we're working with within the AI space are talking about search to their customers already today."

Fortanix's private AI search is pinned on confidential computing, which creates secure vaults that are accessible to authorized parties only via keys. The data is processed inside, without leaving the vault.

"Data is going to be subject to not only the data protection regulations that we have today, but also these emerging AI regulations," Searle says.

Fortanix's technology is a component in the emerging field of confidential AI. Confidential AI is extended to AI scenarios typically associated with GPUs and accelerators, said Mark Russinovich, chief technology officer for Microsoft Azure, during a panel discussion at the Open Confidential Computing Conference in March.

**Protecting Vector Databases**

Private search with AI relies on data extraction from knowledge graphs or vector databases, which could draw information from a wide range of sources, including static conventional databases.

At Nvidia's GPU Technology Conference, also in March, the company's CEO, Jensen Huang, defined vector databases as a new style of database that takes structured data or unstructured data that can be reindexed by encoding the meaning of data.

"Now this becomes an AI database, and that AI database in the future, once you create it, you can talk to it," Huang said.

Fortanix's goal is to initiate privacy for the search initiator — a human or machine user — and protect the privacy and integrity of the information that might be within the vector embeddings.

"Confidential computing, I think, has a very important role to play," Fortanix’s Searle says.

**Private Search Differs From Conventional Search**

Confidential AI prevents the leak of AI information, which helps meet regulatory requirements. The layer also anonymizes the information so a user's intent or identity is protected. That is the opposite of conventional search, where user information and motives drive Google Ads and analytics.

A higher level of AI privacy is a cornerstone for industries such as healthcare and banking, which are tied to regulatory issues.

"Imagine that you want to do search in a scientific domain — you might want to be able to use data from a different institution to the one you're working in that has some specific expertise in a particular field. Showing the privacy of that data and the accuracy of this research is important," Searle says.

**State of Confidential AI**

Confidential AI is an emerging concept, and search fits into that profile. Intel and AMD have chips with hooks for secure enclaves. Microsoft, Google, and Amazon are providing confidential computing virtual machines in their cloud services.

Companies can bring data sets together to train a foundational model effectively, and these data sets are becoming high-value assets for top corporations, said Ian Buck, vice president and general manager of Nvidia's hyperscale and high-performance computing division, during the panel discussion.

The IT industry is rapidly approaching the deployment of confidential computing in data centers, edge computing, and PCs, but there's a lot more to do with confidential AI, said Greg Lavender, Intel's chief technology officer, during the panel.

"Privacy and safety and responsible AI are going to be big forcing functions for this as the government adopts AI technologies from the industry," he said.

**About the Author(s)**

Agam Shah has covered enterprise IT for more than a decade. Outside of machine learning, hardware, and chips, he's also interested in martial arts and Russia.

Fortanix Builds Private Search for AI

The initiative is meant to provide more resources and better strategies for healthcare entities that face an increasing amount of cybersecurity challenges.

Source: Olekcii Mach via Alamy Stock Photo

The Department of Health and Human Services (HHS) has begun an initiative to better organize and equip its healthcare cybersecurity programs through a one-stop shop.

This latest resource is created through the HHS Administration for Strategic Preparedness and Response (ASPR), which leads the US during disasters and public health emergencies relating to health and medical preparation.

This initiative comes after a United Healthcare subsidiary was targeted by BlackCat ransomware group in February, causing days of outages and chaos across the healthcare supply chain. The cyberattack was considered one of the most serious of its kind within the healthcare sector, and led to United paying the ransom demanded by the threat actors. 

The incident prompted a letter from the chairman of the Senate's Homeland Security and Governmental Affairs Committee, Sen. Gary Peters (D-Mich.). He asked HHS what steps it was taking to prevent such a catastrophe from happening again. The ASPR said it will create new cybersecurity strategies for the healthcare sector and inform healthcare entities of best practices and resources.

"There's too many doors into cybersecurity when engaging with the federal government generally, let alone HHS," said Brian Mazanec, the deputy director for ASPR's Office of Preparedness, in a statement. "Within HHS, there are a lot of different players. So we're in the process now of really establishing this front door through ASPR to all of those resources."

**About the Author(s)**

HHS Plans for Cyber 'One-Stop Shop' After United Healthcare Attack

Campaign distributes malware disguised as legitimate installers for popular workplace collaboration apps by abusing a traffic-tracking feature.

Source: Bits and Splits via Shutterstock

Attackers are once again abusing Google Ads to target people with info-stealing malware, this time using an ad-tracking feature to lure corporate users with fake ads for popular collaborative groupware such as Slack and Notion.

Researchers from AhnLab Security Intelligence Center (ASEC) discovered a malicious campaign that uses a statistical feature to embed URLs that distribute malware, including the Rhadamanthys stealer, they revealed in a blog post published this week. The feature lets advertisers insert external analytic website addresses into ads to collect and use their visitors' access-related data to calculate ad traffic.

However, instead of inserting a URL for an external statistics site, attackers are abusing the feature to enter sites for distributing malicious code, the researchers found.

Ads related to the campaign have already been deleted. But when they were still active, "clicking on the banner would take unsuspecting users to the address that would trick them into downloading a malicious file," according to ASEC.

In the campaign, Rhadamanthys is disguised as an installer for popular groupware often used by corporate teams for workplace collaboration. Once the malware is installed and executed, it downloads malicious files and payloads from the attacker's server.

**Redirects to Stealer Downloads**

The ASEC post breaks down how attackers crafted the campaign to show banner ads that contain tracking URLs invisible to the end user that redirect users to an attacker-created and -controlled URL. This ultimate landing page is similar to the actual website of a groupware tool such as Slack or Notion, and it prompts visitors to download and execute the malware, which is distributed in an installer form.

Typical installers used by the campaign are the Inno Setup installer or Nullsoft Scriptable Install System (NSIS) installer; specifically, attackers used the following executable files: Notion\_software\_x64\_.exe Slack\_software\_x64\_.exe; Trello\_software\_x64\_.exe; and GoodNotes\_software\_x64\_32.exe.

"Once it is executed, the malware uses websites that can save texts such as textbin or tinyurl to access the malicious payload addresses," ASEC said in its blog post, which lists the URLs attackers used to fetch these addresses, which are subsequently delivered to users.

The ultimate payload of the campaign is the Rhadamanthys stealer, which gets injected into legitimate Windows files via the "%system32%" path, according to ASEC. This allows the stealer to exfiltrate users' private data without their knowledge, the researchers noted.

Rhadamanthys is popular with attackers and is available for purchase on the Dark Web under a malware-as-a-service model. It acts as a typical stealer to collect system information, such as computer name, username, OS version, and other machine details. It also queries the directories of installed browsers — including Brave, Edge, Chrome, Firefox, Opera Software — to search for and steal browser history, bookmarks, cookies, auto-fills, login credentials, and other data.

**Pay Attention to Ad-Delivered URLs**

The campaign is certainly not the first time that attackers have abused Google Ads and its associated features to deliver Rhadamanthys and other malware, and it likely won't be the last. In fact, a campaign identified in January 2023 also used website redirects from Google Ads and fake-download lures for popular remote-workforce software, such as Zoom and AnyDesk to deliver Rhadamanthys.

Attackers have even abused the "dynamic search ads" feature of the service to amplify the effect of malicious campaigns by creating targeted ads to deliver a flood of malware.

Indeed, as "all search engines that provide tracking to calculate ad traffic can be used to distribute malware," users must stay vigilante when accessing links from ads delivered by Google, ASEC warned. Specifically, they should "pay attention to the URL that is seen upon accessing the website, not the URL that is shown on the ad's banner" to avoid falling for a malicious campaign, according to the post.

ASEC also posted a comprehensive list of URLs associated with various stages of the campaign to help administrators identify if any corporate users have been affected by it.

**About the Author(s)**

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.

Attackers Abuse Google Ad Feature to Target Slack, Notion Users

Our Trusted Advisor dashboard provides an easy-to-understand assessment of your device’s security.

First released for Windows last year, the Malwarebytes Trusted Advisor dashboard is also now available on Mac, iOS and Android. 

Our Trusted Advisor dashboard provides an easy-to-understand assessment of your device’s security, with a single comprehensive protection score, and clear, expert-driven advice. 

In our recent report, “Everyone’s afraid of the internet, and no-one’s sure what to do about it,” we found that only half of the people surveyed feel confident they know how to stay safe online and even fewer are taking the right measures. 

So, though the fears are big, they are followed by very little action. We want to make things easy for our customers so they know what they should be doing, and how. 

Computer security can be difficult and time consuming, especially if you consider all the different devices and operating systems. We want to help our customers, whatever they use. 

Getting it right means knowing what software needs to be updated, whether your system settings are configured securely, and running active protection that can uncover hidden threats. 

Getting it wrong means leaving gaps in your defences that malware, criminal hackers, and other online threats can sneak through. 

Trusted Advisor takes away the guesswork by delivering a holistic assessment of your security and privacy in a way that’s easy to understand, making issues simple to correct. It combines the proven capabilities of Malwarebytes with the knowledge of the brightest industry experts to give you an expert assessment that puts you one step ahead of the cybercrooks. 

**Protection score**

At the heart of Trusted Advisor is a single, easy-to-understand protection score. If you’re rocking a 100% rating then you know you’re crushing it. 

If your score dips below 100%, we’ll explain why, and offer you a checklist of items to improve your security and boost your score. 

Trusted Advisor’s recommendations are practical and jargon-free, so they’re easy to action.

**Six steps to security**

Trusted Advisor monitors various categories of information around security and privacy to assess your overall Protection Score: 

*   **Real-time protection** monitors your device continuously, stopping and removing threats like malware as they appear. It’s vital for keeping you safe from the most destructive threats and the most common methods of infection, so Trusted Advisor will alert you if you aren’t fully protected. 
*   **Software updates** fix the coding flaws that cybercriminals exploit to steal data or put malware on your system. Staying up to date is one of the most important things you can do for your security, so Trusted Advisor has your back here too. 

*   **General settings** covers settings within Malwarebytes, Operating Systems, or your network preferences. Trusted Advisor checks for settings that may not be configured correctly. For example, on iOS it ensures you have defined a passcode for your device and activated web and call protection. 
*   **Device scans** are routine scans that seek out hidden threats on your system. Trusted Advisor will tell you if you get behind and need to run a scan manually. 
*   **Online privacy** helps you take a proactive stance on your privacy by hiding your IP address and blocking third-party ad trackers, making you’re harder to track on the web. Trusted Advisor monitors this so you only part with the personal information you intend to. 
*   **Device health** guards against slowdowns and other performance problems. Trusted Advisor helps you get the most out of your system so that you aren’t left guessing whether it was malware grinding your device to a halt. 

Even with an excellent score, you can’t guarantee absolute safety, though it places you in the closest proximity to it. By following our recommendations, you’ll be in the best security situation you can be.

**Try it today**

If you’re an existing Malwarebytes customer you will get Trusted Advisor automatically, but if you’re in a hurry, you can go to **Settings** > **About** > **Check for updates** and get it right now. If you aren’t, you can get Trusted Advisor by just **downloading the latest version of Malwarebytes**.

 Trusted Advisor now available for Mac, iOS, and Android   

Ransomware groups tore into manufacturing other parts of the OT sector in 2023, and a few attacks caused eight- and nine-figure damages. But worse is yet to come in 2024.

Source: Imagebroker via Alamy Stock Photo

At least 68 cyberattacks last year caused physical consequences to operational technology (OT) networks at more than 500 sites worldwide — in some cases causing $10 million to $100 million in damages.

Unsurprisingly, these weren't Stuxnet-like events, but the opposite.

According to a new report from industrial control system (ICS) vendor Waterfall Security Solutions, which studied real-world cyberattacks on OT organizations, most of the hackers known to be targeting the OT sector these days are hacktivists. And the majority of disruptions are not caused by such direct manipulation of OT systems but are downstream consequences of IT-based attacks, most often involving ransomware.

That doesn't mean, though, that the impacts are any less severe. Incidents involving Johnson Controls and Clorox last year ended up costing those companies around $27 million and $49 million, respectively. One cyberattack that led to the temporary suspension of operations at MKS Instruments in Massachusetts cost $200 million, and one of its suppliers — California-based Applied Materials Inc. — reported losing another $250 million as a result.

The number of attacks with physical consequences increased by nearly 20% last year, according to the report.

**IT Attacks With OT Consequences**

In the past decade and a half, only around a quarter of cyberattacks with OT consequences were caused by actually hitting the OT network, according to the report Waterfall published in collaboration with OT incident threat database ICS STRIVE.

"A large fraction of attacks that caused OT consequences did so by compromising machines in the IT network exclusively," explains Andrew Ginter, vice president of industrial security for Waterfall, and a co-author of the report. "OT was often shut down in 'an abundance of caution' because the business was not willing to keep running powerful, dangerous physical processes with compromise only one or two network hops away."

After its attack last March, for example, the German manufacturer Hahn Group GmbH switched off all of its systems as a safety precaution. A full, clean restoration of its systems took weeks thereafter. A number of other manufacturers last year followed that same playbook, even when safety wasn't at risk, in order to contain damage to further systems, sites, and customers.

"OT was also often shut down because physical operations needed facilities on IT networks that ransomware had crippled — e.g., container-tracking systems for shipping or passenger signage for large rail stations," Ginter points out.

One prime case occurred last January, when UK Royal Mail printers were disabled and hijacked to print LockBit ransom notes. Mail export services were briefly suspended nationwide, in an event that ended up costing £42 million.

"These dependencies are something many OT practitioners do not think about," Ginter explains. An IT network compromise can also affect physical operations, even if an OT network is secured, if the OT process rely on processes in the IT network.

**Cyber Threat to Water Treatment**

More than half of publicly reported cyberattacks with OT consequences in 2023 affected the manufacturing sector. But if there's one sector to worry about more than the rest it's, arguably, water.

Late last November, around 180 households in the Irish villages of Binghamstown and Drum lost water for two days, thanks to a loss of water pressure at a local pumping station. The cause was a cyberattack likely carried out by Iran's Cyber Av3ngers, part of a wider campaign targeting Unitronics pump controllers.

Though such stories are still rare, water facilities combine a dangerous mix of low difficulty and high impact for hackers.

"In the USA, the vast majority of the more than 20,000 drinking water treatment utilities are tiny. Minute. The vast majority of the more than 200,000 wastewater treatment systems — same thing. And realistically, with whatever budget these utilities have, almost all of it goes to people with trucks and backhoes digging holes in the ground," Ginter explains. "Couple that with continued pressure to automate those water systems to reduce costs — a lot of these systems are regulated \[because they're local monopolies\], and every regulator wants to reduce costs and reduce rates, so there is constant pressure to automate. All modern automation involves computers, meaning more targets for cyberattacks."

These systems have no security budget, so with the increased threat of hacktivist attacks and pressure to automate their operations, they are in peril, he notes, creating "a growing problem for all of the small communities in the nation."

**About the Author(s)**

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.

Cyberattacks Wreaking Physical Disruption on the Rise

By Waqas
Critical Backdoor Alert! Patch XZ Utils Now (CVE-2024-3094) & Secure Your Linux System. Learn how a hidden backdoor…
This is a post from HackRead.com Read the original post: Backdoor Discovered in XZ Utils: Patch Your Systems Now (CVE-2024-3094)

Critical Backdoor Alert! Patch XZ Utils Now (CVE-2024-3094) & Secure Your Linux System. Learn how a hidden backdoor puts Linux at risk and how to patch it immediately.

A critical security vulnerability, designated CVE-2024-3094, was recently discovered in the widely used XZ Utils package. This vulnerability threatens Linux systems with backdoor attacks.

For your information, XZ Utils is a collection of open-source command-line tools for data compression and decompression. It includes the popular xz command and the liblzma library, which is used by other software, most notably OpenSSH – the program that enables secure remote access to Linux systems.

****The Backdoor Explained****

The vulnerability involved a malicious backdoor hidden within the source code of XZ Utils, specifically in the liblzma library. This backdoor code, if triggered, could allow an attacker to gain unauthorized remote access to a vulnerable system through SSH. The attacker wouldn’t even need valid credentials, potentially granting complete control over the system.

****Impact and Discovery****

The potential impact of this vulnerability is severe. An attacker exploiting CVE-2024-3094 could steal sensitive data, install malware, disrupt critical operations, or even use the compromised system to launch further attacks.

Fortunately, the backdoor was discovered by the security community in late March 2024 before widespread distribution. This prevented a large-scale security breach. However, some Linux users remain vulnerable, especially those using unstable or rolling-release distributions.

****Who is Affected?****

According to OpenSSH’s report, the specific versions of XZ Utils containing the backdoor were 5.6.0 and 5.6.1. These versions were only recently released and did not make it into the stable branches of most major Linux distributions. However, users who manually compiled these versions from source code or installed them from non-standard repositories could be at risk.

Commenting on this, John Bambenek, President at Bambenek Consulting warned, _“_The original reports of this backdoor showed exploitation of this vulnerability via SSH which means it can be triggered even if the victim machine’s users don’t use XZ and its library. It seems this library tends to be installed by default on modern Linux distributions so organizations should immediately prioritize downgrading the package until a safe update is released, even if they don’t use the tools themselves._“_

****Mitigation and Prevention****

The most critical step to address this vulnerability is to update your system immediately. Most Linux distributions have released patch updates for XZ Utils. Here’s how to update depending on your distribution:

*   **Debian/Ubuntu:** Use sudo apt update and sudo apt upgrade commands.
*   **Red Hat/CentOS/Fedora:** Use sudo dnf update command.
*   **Other Distributions:** Refer to your distribution’s specific update instructions.

****Lessons Learned****

The discovery of CVE-2024-3094 emphasizes the importance of various security measures. Firstly, keeping software and systems updated with regular patches is crucial to mitigate potential risks.

Secondly, maintaining a sharp review process for open-source projects aids in the early detection of vulnerabilities. Thirdly, promoting security awareness among users and employees through education about risks and best practices is essential for enhancing overall protection. Adhering to these practices enables us to reduce the impact of vulnerabilities like CVE-2024-3094 and safeguard the security of our systems.

1.  New Linux Malware Alert: ‘Spinning YARN’ Hits Docker
2.  Crypto Stealing PyPI Malware Hits Windows, Linux Users
3.  Magnet Goblin Using Ivanti Flaws to Deploy Linux Malware
4.  Bifrost RAT Variant Hits Linux Devices, Mimics VMware Domain
5.  Xamalicious Backdoor Infects Android Apps, Affects 327K Devices

Backdoor Discovered in XZ Utils: Patch Your Systems Now (CVE-2024-3094)

Microsoft adds tools to protect Azure AI from threats such as prompt injection, as well as to give developers the capabilities to ensure generative AI apps are more resilient to model and content manipulation attacks.

Source: Tada Images via Shutterstock

Microsoft has announced several new capabilities in Azure AI Studio that the company says should help developers build generative artificial intelligence (GenAI) apps that are more reliable and resilient against malicious model manipulation and other emerging threats.

In a March 29 blog post, Microsoft's chief product officer of responsible AI, Sarah Bird, pointed to growing concerns about threat actors using prompt injection attacks to get AI systems to behave in dangerous and unexpected ways as the primary driving factor for the new tools.

"Organizations are also concerned about quality and reliability," Bird said. "They want to ensure that their AI systems are not generating errors or adding information that isn’t substantiated in the application’s data sources, which can erode user trust."

Azure AI Studio is a hosted platform that organizations can use to build custom AI assistants, copilots, bots, search tools and other applications, grounded in their own data. Announced in November, the platform hosts Microsoft's machine learning models, as well as models from several other sources, including OpenAI, Meta, Hugging Face, and Nvidia. It allows developers to quickly integrate multimodal capabilities and responsible AI features into their models.

Other major players, such as Amazon and Google, have rushed to market with similar offerings over the past year to tap into the surging interest in AI technologies worldwide. A recent IBM-commissioned study found that 42% of organizations with more than 1,000 employees are already actively using AI in some fashion, with many of them planning to increase and accelerate investments in the technology over the next few years. And not all of them were telling IT beforehand about their AI use.

**Protecting Against Prompt Engineering**

The five new capabilities that Microsoft has added — or will soon add — to Azure AI Studio are Prompt Shields, groundedness detection, safety system messages, safety evaluations, and risk and safety monitoring. The features are designed to address some significant challenges that researchers have uncovered recently — and continue to uncover on a routine basis — with regard to the use of large language models (LLMs) and GenAI tools.

Prompt Shields, for instance, is Microsoft's mitigation for what are known as indirect prompt attacks and jailbreaks. The feature builds on existing mitigations in Azure AI Studio against jailbreak risk. In prompt-engineering attacks, adversaries use prompts that appear innocuous and not overtly harmful to try and steer an AI model into generating harmful and undesirable responses. Prompt engineering is among the most dangerous in a growing class of attacks that try to jailbreak AI models or get them to behave in a manner that is inconsistent with any filters and constraints that developers might have built into them.  

Researchers have recently shown how adversaries can engage in prompt-engineering attacks to get GenAI models to spill their training data, spew out personal information, generate misinformation, and potentially harmful content, such as instructions on how to hot-wire a car.

With Prompt Shields, developers can integrate capabilities into their models that help distinguish between valid and potentially untrustworthy system inputs, set delimiters to help mark the beginning and end of input text, and use data marking to mark input texts. Prompt Shields is currently available in preview mode in Azure AI Content Safety and will become generally available soon, according to Microsoft.

**Mitigations for Model Hallucinations and Harmful Content**

With groundedness detection, Microsoft has added a feature to Azure AI Studio that it says can help developers reduce the risk of their AI models "hallucinating." Model hallucination is a tendency by AI models to generate results that appear plausible but are completely made up and not based — or grounded — on the training data. LLM hallucinations can be hugely problematic if an organization were to take the output as factual and act on it in some way. In a software development environment, for instance, LLM hallucinations could result in developers potentially introducing vulnerable code into their applications.

Azure AI Studio's new groundedness-detection capability is basically about helping detect — more reliably and at greater scale — potentially ungrounded GenAI outputs.  The goal is to give developers a way to test their AI models against what Microsoft calls "groundedness metrics" before deploying the model into their products. The feature also highlights potentially ungrounded statements in LLM outputs, so users know to fact check the output before using it. Groundedness detection should be available in the near future, according to Microsoft.

The new system message framework offers a way for developers to clearly define their models' capabilities, profile, and limitations in their specific environments. Developers can use the capability to define the format of the output and provide examples of intended behavior, so it becomes easier for users to detect deviations from intended behavior. The feature isn't available yet but should be soon.

Azure AI Studio's newly announced safety evaluations capability and its risk and safety monitoring feature are both currently available in preview status. Organizations can use the former to assess the vulnerability of their LLM models to jailbreak attacks and generate unexpected content. The risk and safety monitoring capability allows developers to detect model inputs that are problematic and likely to trigger hallucinated or unexpected content, so they can implement mitigations against it.

"Generative AI can be a force multiplier for every department, company, and industry," Microsoft's Bird said. "At the same time, foundation models introduce new challenges for security and safety that require novel mitigations and continuous learning."

**About the Author(s)**

Jai Vijayan is a seasoned technology reporter with over 20 years of experience in IT trade journalism. He was most recently a Senior Editor at Computerworld, where he covered information security and data privacy issues for the publication. Over the course of his 20-year career at Computerworld, Jai also covered a variety of other technology topics, including big data, Hadoop, Internet of Things, e-voting, and data analytics. Prior to Computerworld, Jai covered technology issues for The Economic Times in Bangalore, India. Jai has a Master's degree in Statistics and lives in Naperville, Ill.

Microsoft Beefs Up Defenses in Azure AI

From ransomware to LOTL attacks,  the top threats your educational institution needs to be aware of in 2024.

Educational institutions may face a range of cyberthreats in 2024, but our 2024 State of Malware in Education report identifies the six most critical ones.

Ransomware, for example, stands out as a key threat for schools and universities. The report covers how last year, we witnessed a 92% increase in ransomware attacks in K-12 schools and a 70% increase in Higher Education. The trend appears set to continue, partly due to specialized ransomware groups like Rhysida (formerly Vice Society) targeting educational sectors.  

**Education ransomware attacks, 2022 – 2023**

Another major threat our 2024 State of Malware in Education covers is the reduction of conventional malware in favor of Living off The Land (LOTL) attacks. LOTL attacks exploit legitimate system tools to remain undetected while conducting harmful activities.

Our report suggests that educational institutions must employ expert staff to manually identify LOTL activities, which traditional malware detection tools miss. For example, we recently wrote how one K-12 district used MDR to uncover malicious PowerShell activity and stop an ongoing infection.

Some other trends and threats educational institutions can expect in the report to cover include:

*   Why targeting Macs has become **an easy choice for criminals** 
*   How CL0P is r**ewriting the ransomware playbook** and why Big Game ransomware remains the most serious threat.
*   How cybercriminals use ‘malvertising’ to **target educational institutions with malicious ads** for popular for remote learning such as Zoom. 

As we progress into 2024, the reality is that educational institutions’ success in pairing state of the art security software with skilled security staff will be a deciding factor in their ability to take down the most serious cyberthreats. 

To understand the complete list of threats facing educational institutions in 2024 and how to tackle them, get the full 2024 State of Malware in Education report—tailored to either K-12 or Higher Ed—below.

 2024 State of Malware in Education report: Top 6 cyberthreats facing K-12 and Higher Ed 

Gentoo Linux Security Advisory 202403-4 - A backdoor has been discovered in XZ utils that could lead to remote compromise of systems. Versions less than 5.6.0 are affected.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -  
Gentoo Linux Security Advisory                           GLSA 202403-04  
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -  
                                           https://security.gentoo.org/  
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

 Severity: High  
    Title: XZ utils: Backdoor in release tarballs  
     Date: March 29, 2024  
     Bugs: #928134  
       ID: 202403-04

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Synopsis  
========

A backdoor has been discovered in XZ utils that could lead to remote  
compromise of systems.

Background  
==========

XZ Utils is free general-purpose data compression software with a high  
compression ratio.

Affected packages  
=================

Package            Vulnerable    Unaffected  
-----------------  ------------  ------------  
app-arch/xz-utils  >= 5.6.0      < 5.6.0

Description  
===========

A backdoor has been discovered in XZ utils. Please review the CVE  
identifier referenced below for details.

Impact  
======

Our current understanding of the backdoor is that is does not affect  
Gentoo systems, because

1. the backdoor only appears to be included on specific systems and  
Gentoo does not qualify;  
2. the backdoor as it is currently understood targets OpenSSH patched to  
work with systemd-notify support. Gentoo does not support or include  
these patches;

Analysis is still ongoing, however, and additional vectors may still be  
identified. For this reason we are still issuing this advisory as if  
that will be the case.

Workaround  
==========

There is no known workaround at this time.

Resolution  
==========

All XZ utils users should downgrade to the latest version before the  
backdoor was introduced:

  # emerge --sync  
  # emerge --ask --oneshot --verbose "<app-arch/xz-utils-5.6.0"

References  
==========

[ 1 ] CVE-2024-3094  
      https://nvd.nist.gov/vuln/detail/CVE-2024-3094

Availability  
============

This GLSA and any updates to it are available for viewing at  
the Gentoo Security Website:

 https://security.gentoo.org/glsa/202403-04

Concerns?  
=========

Security is a primary focus of Gentoo Linux and ensuring the  
confidentiality and security of our users' machines is of utmost  
importance to us. Any security concerns should be addressed to  
security@gentoo.org or alternatively, you may file a bug at  
https://bugs.gentoo.org.

License  
=======

Copyright 2024 Gentoo Foundation, Inc; referenced text  
belongs to its owner(s).

The contents of this document are licensed under the  
Creative Commons - Attribution / Share Alike license.

https://creativecommons.org/licenses/by-sa/2.5

Tag

#mac