Apr 12, 2024NewsroomCyber Attack / Data Breach

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Thursday issued an emergency directive (ED 24-02) urging federal agencies to hunt for signs of compromise and enact preventive measures following the recent compromise of Microsoft’s systems that led to the theft of email correspondence with the company.

The attack, which came to light earlier this year, has been attributed to a Russian nation-state group tracked as Midnight Blizzard (aka APT29 or Cozy Bear). Last month, Microsoft revealed that the adversary managed to access some of its source code repositories but noted that there is no evidence of a breach of customer-facing systems.

The emergency directive, which was originally issued privately to federal agencies on April 2, was first reported on by CyberScoop two days later.

“The threat actor is using information initially exfiltrated from the corporate email systems, including authentication details shared between Microsoft customers and Microsoft by email, to gain, or attempt to gain, additional access to Microsoft customer systems,” CISA said.

The agency said the theft of email correspondence between government entities and Microsoft poses severe risks, urging concerned parties to analyze the content of exfiltrated emails, reset compromised credentials, and take additional steps to ensure authentication tools for privileged Microsoft Azure accounts are secure.

It’s currently not clear how many federal agencies have had their email exchanges exfiltrated in the wake of the incident, although CISA said all of them have been notified.

The agency is also urging affected entities to perform a cybersecurity impact analysis by April 30, 2024, and provide a status update by May 1, 2024, 11:59 p.m. Other organizations that are impacted by the breach are advised to contact their respective Microsoft account team for any additional questions or follow up.

“Regardless of direct impact, all organizations are strongly encouraged to apply stringent security measures, including strong passwords, multi-factor authentication (MFA) and prohibited sharing of unprotected sensitive information via unsecure channels,” CISA said.

The development comes as CISA released a new version of its malware analysis system, called Malware Next-Gen, that allows organizations to submit malware samples (anonymously or otherwise) and other suspicious artifacts for analysis.

Microsoft has released security updates for the month of April 2024 to remediate a record 149 flaws, two of which have come under active exploitation in the wild.

Of the 149 flaws, three are rated Critical, 142 are rated Important, three are rated Moderate, and one is rated Low in severity. The update is aside from 21 vulnerabilities that the company addressed in its Chromium-based Edge browser following the release of the March 2024 Patch Tuesday fixes.

The two shortcomings that have come under active exploitation are below –

• CVE-2024-26234 (CVSS score: 6.7) – Proxy Driver Spoofing Vulnerability
• CVE-2024-29988 (CVSS score: 8.8) – SmartScreen Prompt Security Feature Bypass Vulnerability

While Microsoft’s own advisory provides no information about CVE-2024-26234, cybersecurity firm Sophos said it discovered in December 2023 a malicious executable (“Catalog.exe” or “Catalog Authentication Client Service”) that’s signed by a valid Microsoft Windows Hardware Compatibility Publisher (WHCP) certificate.

Authenticode analysis of the binary has revealed the original requesting publisher to Hainan YouHu Technology Co. Ltd, which is also the publisher of another tool called LaiXi Android Screen Mirroring.

The latter is described as “a marketing software … [that] can connect hundreds of mobile phones and control them in batches, and automate tasks like batch following, liking, and commenting.”

Present within the purported authentication service is a component called 3proxy that’s designed to monitor and intercept network traffic on an infected system, effectively acting as a backdoor.

“We have no evidence to suggest that the LaiXi developers deliberately embedded the malicious file into their product, or that a threat actor conducted a supply chain attack to insert it into the compilation/building process of the LaiXi application,” Sophos researcher Andreas Klopsch said.

The cybersecurity company also said it discovered multiple other variants of the backdoor in the wild going all the way back to January 5, 2023, indicating that the campaign has been underway at least since then. Microsoft has since added the relevant files to its revocation list.

The other security flaw that has reportedly come under active attack is CVE-2024-29988, which – like CVE-2024-21412 and CVE-2023-36025 – allows attackers to sidestep Microsoft Defender Smartscreen protections when opening a specially crafted file.

“To exploit this security feature bypass vulnerability, an attacker would need to convince a user to launch malicious files using a launcher application that requests that no UI be shown,” Microsoft said.

“In an email or instant message attack scenario, the attacker could send the targeted user a specially crafted file that is designed to exploit the remote code execution vulnerability.”

The Zero Day Initiative revealed that there is evidence of the flaw being exploited in the wild, although Microsoft has tagged it with an “Exploitation More Likely” assessment.

Another vulnerability of importance is CVE-2024-29990 (CVSS score: 9.0), an elevation of privilege flaw impacting Microsoft Azure Kubernetes Service Confidential Container that could be exploited by unauthenticated attackers to steal credentials.

“An attacker can access the untrusted AKS Kubernetes node and AKS Confidential Container to take over confidential guests and containers beyond the network stack it might be bound to,” Redmond said.

In all, the release is notable for addressing as many as 68 remote code execution, 31 privilege escalation, 26 security feature bypass, and six denial-of-service (DoS) bugs. Interestingly, 24 of the 26 security bypass flaws are related to Secure Boot.

“While none of these Secure Boot vulnerabilities addressed this month were exploited in the wild, they serve as a reminder that flaws in Secure Boot persist, and we could see more malicious activity related to Secure Boot in the future,” Satnam Narang, senior staff research engineer at Tenable, said in a statement.

The disclosure comes as Microsoft has faced criticism for its security practices, with a recent report from the U.S. Cyber Safety Review Board (CSRB) calling out the company for not doing enough to prevent a cyber espionage campaign orchestrated by a Chinese threat actor tracked as Storm-0558 last year.

It also follows the company’s decision to publish root cause data for security flaws using the Common Weakness Enumeration (CWE) industry standard. However, it’s worth noting that the changes are only in effect starting from advisories published since March 2024.

“The addition of CWE assessments to Microsoft security advisories helps pinpoint the generic root cause of a vulnerability,” Adam Barnett, lead software engineer at Rapid7, said in a statement shared with The Hacker News.

“The CWE program has recently updated its guidance on mapping CVEs to a CWE Root Cause. Analysis of CWE trends can help developers reduce future occurrences through improved Software Development Life Cycle (SDLC) workflows and testing, as well as helping defenders understand where to direct defense-in-depth and deployment-hardening efforts for best return on investment.”

In a related development, cybersecurity firm Varonis detailed two methods that attackers could adopt to circumvent audit logs and avoid triggering download events while exfiltrating files from SharePoint.

The first approach takes advantage of SharePoint’s “Open in App” feature to access and download files, whereas the second uses the User-Agent for Microsoft SkyDriveSync to download files or even entire sites while miscategorizing such events as file syncs instead of downloads.

Microsoft, which was made aware of the issues in November 2023, has yet to release a fix, although they have been added to their patch backlog program. In the interim, organizations are recommended to closely monitor their audit logs for suspicious access events, specifically those that involve large volumes of file downloads within a short period.

“These techniques can bypass the detection and enforcement policies of traditional tools, such as cloud access security brokers, data loss prevention, and SIEMs, by hiding downloads as less suspicious access and sync events,” Eric Saraga said.

Software Patches from Other Vendors

In addition to Microsoft, security updates have also been released by other vendors over the past few weeks to rectify several vulnerabilities, including —

Apr 03, 2024NewsroomData Breach / Incident Response

The U.S. Cyber Safety Review Board (CSRB) has criticized Microsoft for a series of security lapses that led to the breach of nearly two dozen companies across Europe and the U.S. by a China-based nation-state group called Storm-0558 last year.

The findings, released by the Department of Homeland Security (DHS) on Tuesday, found that the intrusion was preventable, and that it became successful due to a “cascade of Microsoft’s avoidable errors.”

“It identified a series of Microsoft operational and strategic decisions that collectively pointed to a corporate culture that deprioritized enterprise security investments and rigorous risk management, at odds with the company’s centrality in the technology ecosystem and the level of trust customers place in the company to protect their data and operations,” the DHS said in a statement.

The CSRB also lambasted the tech titan for failing to detect the compromise on its own, instead relying on a customer to reach out to flag the breach. It further faulted Microsoft for not prioritizing the development of an automated key rotation solution and rearchitecting its legacy infrastructure to meet the needs of the current threat landscape.

The incident first came to light in July 2023 when Microsoft revealed that Storm-0558 gained unauthorized access to 22 organizations as well as more than more than 500 related individual consumer accounts.

Microsoft subsequently said a validation error in its source code made it possible for Azure Active Directory (Azure AD) tokens to be forged by Storm-0558 using a Microsoft account (MSA) consumer signing key, thus allowing the adversary to infiltrate the mailboxes.

In September 2023, the company divulged that Storm-0558 acquired the consumer signing key to forge the tokens by compromising an engineer’s corporate account that had access to a debugging environment hosting a crash dump of its consumer signing system that also inadvertently contained the signing key.

Microsoft has since acknowledged in a March 2024 update that it was inaccurate and that it has not still been able to locate a “crash dump containing the impacted key material.” It also said its investigation into the hack remains ongoing.

“Our leading hypothesis remains that operational errors resulted in key material leaving the secure token signing environment that was subsequently accessed in a debugging environment via a compromised engineering account,” it noted.

“Recent events have demonstrated a need to adopt a new culture of engineering security in our own networks,” a Microsoft spokesperson was quoted as saying to The Washington Post.

As many as 60,000 unclassified emails from Outlook accounts are believed to have been exfiltrated over the course of the campaign that began in May 2023. China has rejected accusations that it was behind the attack.

Earlier this February, Redmond expanded free logging capabilities to all U.S. federal agencies using Microsoft Purview Audit, irrespective of the license tier, to help them detect, respond, and prevent sophisticated cyber attacks.

“The threat actor responsible for this brazen intrusion has been tracked by industry for over two decades and has been linked to 2009 Operation Aurora and 2011 RSA SecureID compromises,” said CSRB Acting Deputy Chair Dmitri Alperovitch.

“This People’s Republic of China affiliated group of hackers has the capability and intent to compromise identity systems to access sensitive data, including emails of individuals of interest to the Chinese government.”

To safeguard against threats from state-sponsored actors, cloud service providers have been recommended to –

• Implement modern control mechanisms and baseline practices
• Adopt a minimum standard for default audit logging in cloud services
• Incorporate emerging digital identity standards to secure cloud services
• Adopt incident and vulnerability disclosure practices to maximize transparency
• Develop more effective victim notification and support mechanisms to drive information-sharing efforts

“The United States government should update the Federal Risk Authorization Management Program and supporting frameworks and establish a process for conducting discretionary special reviews of the program’s authorized Cloud Service Offerings following especially high-impact situations,” the CSRB said.

Mar 27, 2024NewsroomThreat Intelligence / Network Security

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added a security flaw impacting the Microsoft Sharepoint Server to its Known Exploited Vulnerabilities (KEV) catalog based on evidence of active exploitation in the wild.

The vulnerability, tracked as CVE-2023-24955 (CVSS score: 7.2), is a critical remote code execution flaw that allows an authenticated attacker with Site Owner privileges to execute arbitrary code.

“In a network-based attack, an authenticated attacker as a Site Owner could execute code remotely on the SharePoint Server,” Microsoft said in an advisory. The flaw was addressed by Microsoft as part of its Patch Tuesday updates for May 2023.

The development comes more than two months after CISA added CVE-2023-29357, a privilege escalation flaw in SharePoint Server, to its KEV catalog.

It’s worth pointing out that an exploit chain combining CVE-2023-29357 and CVE-2023-24955 was demonstrated by StarLabs SG at the Pwn2Own Vancouver hacking contest last year, earning the researchers a $100,000 prize.

That said, there is currently no information on the attacks weaponizing these two vulnerabilities and the threat actors that may be exploiting them.

Microsoft previously told The Hacker News that “customers who have enabled automatic updates and enable ‘Receive updates for other Microsoft products’ option within their Windows Update settings are already protected.”

Federal Civilian Executive Branch (FCEB) agencies are required to apply the fixes by April 16, 2024, to secure their networks against active threats.

Mar 27, 2024NewsroomVulnerability / API Security

A now-patched security flaw in the Microsoft Edge web browser could have been abused to install arbitrary extensions on users’ systems and carry out malicious actions.

“This flaw could have allowed an attacker to exploit a private API, initially intended for marketing purposes, to covertly install additional browser extensions with broad permissions without the user’s knowledge,” Guardio Labs security researcher Oleg Zaytsev said in a new report shared with The Hacker News.

Tracked as CVE-2024-21388 (CVSS score: 6.5), it was addressed by Microsoft in Edge stable version 121.0.2277.83 released on January 25, 2024, following responsible disclosure in November 2023. The Windows maker credited both Zaytsev and Jun Kokatsu for reporting the issue.

“An attacker who successfully exploited this vulnerability could gain the privileges needed to install an extension,” Microsoft said in an advisory for the flaw, adding it “could lead to a browser sandbox escape.”

Describing it as a privilege escalation flaw, the tech giant also emphasized that a successful exploitation of the bug requires an attacker to “take additional actions prior to exploitation to prepare the target environment.”

According to Guardio’s findings, CVE-2024-21388 allows a bad actor with the ability to run JavaScript on bing[.]com or microsoft[.]com pages to install any extensions from the Edge Add-ons store sans requiring user’s consent or interaction.

This is made possible by the fact that the browser comes with privileged access to certain private APIs that make it possible to install an add-on as long as it’s from the vendor’s own extension marketplace.

One such API in the Chromium-based Edge browser is edgeMarketingPagePrivate, which is accessible from a set of allowlisted websites that belong to Microsoft, including bing[.]com, microsoft[.]com, microsoftedgewelcome.microsoft[.]com, and microsoftedgetips.microsoft[.]com, among others.

The API also packs in a method called installTheme() that, as the name implies, is designed to install a theme from the Edge Add-ons store by passing a unique theme identifier (“themeId”) and its manifest file as input.

The bug identified by Guardio is essentially a case of insufficient validation, thereby enabling an attacker to provide any extension identifier from the storefront (as opposed to the themeId) and get it stealthily installed.

“As an added bonus, as this extension installation is not done quite in the manner it was originally designed for, there will be no need for any interaction or consent from the user,” Zaytsev explained.

In a hypothetical attack scenario leveraging CVE-2024-21388, a threat actor could publish a seemingly harmless extension to the add-ons store and use it to inject a piece of malicious JavaScript code into bing[.]com – or any of the sites that are allowed to access the API – and install an arbitrary extension of their choice by invoking the API using the extension identifier.

Put differently, executing the specially crafted extension on the Edge browser and going to bing[.]com will automatically install the targeted extension without the victim’s permission.

Guardio told The Hacker News that while there is no evidence of this bug being exploited in the wild, it highlights the need for balancing user convenience and security, and how browser customizations can inadvertently defeat security mechanisms and introduce several new attack vectors.

“It’s relatively easy for attackers to trick users into installing an extension that appears harmless, not realizing it serves as the initial step in a more complex attack,” Zaytsev said. “This vulnerability could be exploited to facilitate the installation of additional extensions, potentially for monetary gain.”

Mar 19, 2024NewsroomSocial Engineering / Email Security

A new phishing campaign is targeting U.S. organizations with the intent to deploy a remote access trojan called NetSupport RAT.

Israeli cybersecurity company Perception Point is tracking the activity under the moniker Operation PhantomBlu.

“The PhantomBlu operation introduces a nuanced exploitation method, diverging from NetSupport RAT’s typical delivery mechanism by leveraging OLE (Object Linking and Embedding) template manipulation, exploiting Microsoft Office document templates to execute malicious code while evading detection,” security researcher Ariel Davidpur said.

NetSupport RAT is a malicious offshoot of a legitimate remote desktop tool known as NetSupport Manager, allowing threat actors to conduct a spectrum of data gathering actions on a compromised endpoint.

The starting point is a Salary-themed phishing email that purports to be from the accounting department and urges recipients to open the attached Microsoft Word document to view the “monthly salary report.”

A closer analysis of the email message headers – particularly the Return-Path and Message-ID fields – shows that the attackers use a legitimate email marketing platform called Brevo (formerly Sendinblue) to send the emails.

The Word document, upon opening, instructs the victim to enter a password provided in the email body and enable editing, followed by double-clicking a printer icon embedded in the doc to view the salary graph.

Doing so opens a ZIP archive file (“Chart20072007.zip”) containing one Windows shortcut file, which functions as a PowerShell dropper to retrieve and execute a NetSupport RAT binary from a remote server.

“By using encrypted .docs to deliver the NetSupport RAT via OLE template and template injection, PhantomBlu marks a departure from the conventional TTPs commonly associated with NetSupport RAT deployments,” Davidpur said, adding the updated technique “showcases PhantomBlu’s innovation in blending sophisticated evasion tactics with social engineering.”

Growing Abuse of Cloud Platforms and Popular CDNs

The development comes as Resecurity revealed that threat actors are increasingly abusing public cloud services like Dropbox, GitHub, IBM Cloud, and Oracle Cloud Storage, as well as Web 3.0 data-hosting platforms built on the InterPlanetary File System (IPFS) protocol such as Pinata to generate fully undetectable (FUD) phishing URLs using phishing kits.

Such FUD links are offered on Telegram by underground vendors like BulletProofLink, FUDLINKSHOP, FUDSENDER, ONNX, and XPLOITRVERIFIER for prices starting at $200 per month as part of a subscription model. These links are further secured behind antibot barriers to filter incoming traffic and evade detection.

Also complementing these services are tools like HeartSender that make it possible to distribute the generated FUD links at scale. The Telegram group associated with HeartSender has nearly 13,000 subscribers.

“FUD Links represent the next step in [phishing-as-a-service] and malware-deployment innovation,” the company said, noting attackers are “repurposing high-reputation infrastructure for malicious use cases.”

“One recent malicious campaign, which leveraged the Rhadamanthys Stealer to target the oil and gas sector, used an embedded URL that exploited an open redirect on legitimate domains, primarily Google Maps and Google Images. This domain-nesting technique makes malicious URLs less noticeable and more likely to entrap victims.”

Mar 14, 2024NewsroomMalware / Cyber Attack

A DarkGate malware campaign observed in mid-January 2024 leveraged a recently patched security flaw in Microsoft Windows as a zero-day using bogus software installers.

“During this campaign, users were lured using PDFs that contained Google DoubleClick Digital Marketing (DDM) open redirects that led unsuspecting victims to compromised sites hosting the Microsoft Windows SmartScreen bypass CVE-2024-21412 that led to malicious Microsoft (.MSI) installers,” Trend Micro said.

CVE-2024-21412 (CVSS score: 8.1) concerns an internet shortcut files security feature bypass vulnerability that permits an unauthenticated attacker to circumvent SmartScreen protections by tricking a victim into clicking on a specially crafted file.

It was fixed by Microsoft as part of its Patch Tuesday updates for February 2024, but not before it was weaponized by a threat actor called Water Hydra (aka DarkCasino) to deliver the DarkMe malware in attacks targeting financial institutions.

The latest findings from Trend Micro show that the vulnerability has come under broader exploitation than previously thought, with the DarkGate campaign leveraging it in conjunction with open redirects from Google Ads to proliferate the malware.

The sophisticated attack chain begins with victims clicking on a link embedded within a PDF attachment sent via a phishing email. The link deploys an open redirect from Google’s doubleclick[.]net domain to a compromised web server hosting a malicious .URL internet shortcut file that exploits CVE-2024-21412.

Specifically, the open redirects are designed to distribute fake Microsoft software installers (.MSI) masquerading as legitimate software, such as Apple iTunes, Notion, NVIDIA, which come fitted with a side-loaded DLL file that decrypted and infected users with DarkGate (version 6.1.7).

It’s worth noting that another now-fixed bypass flaw in Windows SmartScreen (CVE-2023-36025, CVSS score: 8.8) has been employed by threat actors to deliver DarkGate, Phemedrone Stealer, and Mispadu over the past few months.

The abuse of Google Ads technologies allows threat actors to increase the reach and scale of their attacks through different ad campaigns that are tailored for specific audiences.

“Using fake software installers, along with open redirects, is a potent combination and can lead to many infections,” security researchers Peter Girnus, Aliakbar Zahravi, and Simon Zuckerbraun said. “It is essential to remain vigilant and to instruct users not to trust any software installer that they receive outside of official channels.”

The development comes as the AhnLab Security Intelligence Center (ASEC) and eSentire revealed that counterfeit installers for Adobe Reader, Notion and Synaptics are being distributed via fake PDF files and seemingly legitimate websites to deploy information stealers like LummaC2 and the XRed backdoor.

It also follows the discovery of new stealer malware families like Planet Stealer, Rage Stealer (aka xStealer), and Tweaks (aka Tweaker), adding to the plethora of cyber threats that are capable of harvesting sensitive information from compromised hosts.

“Attackers are exploiting popular platforms, like YouTube and Discord, to distribute Tweaks to Roblox users, capitalizing on the ability of legitimate platforms to evade detection by web filter block lists that typically block known malicious servers,” Zscaler ThreatLabz said.

“Attackers share malicious files disguised as Frames Per Second (FPS) optimization packages with users and, in turn, users infect their own systems with Tweaks malware.”

The PowerShell-based stealer is equipped to exfiltrate sensitive data, including user information, location, Wi-Fi profiles, passwords, Roblox IDs, and in-game currency details, to an attacker-controlled server via a Discord webhook.

Malvertising and social engineering campaigns have also been observed acting as an initial access vector to disseminate a wide range of stealer and remote access trojans like Agent Tesla, CyberGate RAT, Fenix botnet, Matanbuchus, NarniaRAT, Remcos RAT, Rhadamanthys, SapphireStealer, and zgRAT.

Mar 09, 2024NewsroomCyber Attack / Threat Intelligence

Microsoft on Friday revealed that the Kremlin-backed threat actor known as Midnight Blizzard (aka APT29 or Cozy Bear) managed to gain access to some of its source code repositories and internal systems following a hack that came to light in January 2024.

“In recent weeks, we have seen evidence that Midnight Blizzard is using information initially exfiltrated from our corporate email systems to gain, or attempt to gain, unauthorized access,” the tech giant said.

“This has included access to some of the company’s source code repositories and internal systems. To date we have found no evidence that Microsoft-hosted customer-facing systems have been compromised.”

Redmond, which is continuing to investigate the extent of the breach, said the Russian state-sponsored threat actor is attempting to leverage the different types of secrets it found, including those that were shared between customers and Microsoft in email.

It, however, did not disclose what these secrets were or the scale of the compromise, although it said it has directly reached out to impacted customers. It’s not clear what source code was accessed.

Stating that it has increased in its security investments, Microsoft further noted that the adversary ramped up its password spray attacks by as much as 10-fold in February, compared to the “already large volume” observed in January.

“Midnight Blizzard’s ongoing attack is characterized by a sustained, significant commitment of the threat actor’s resources, coordination, and focus,” it said.

“It may be using the information it has obtained to accumulate a picture of areas to attack and enhance its ability to do so. This reflects what has become more broadly an unprecedented global threat landscape, especially in terms of sophisticated nation-state attacks.”

The Microsoft breach is said to have taken place in November 2023, with Midnight Blizzard employing a password spray attack to successfully infiltrate a legacy, non-production test tenant account that did not have multi-factor authentication (MFA) enabled.

The tech giant, in late January, revealed that APT29 had targeted other organizations by taking advantage of a diverse set of initial access methods ranging from stolen credentials to supply chain attacks.

Midnight Blizzard is considered part of Russia’s Foreign Intelligence Service (SVR). Active since at least 2008, the threat actor is one of the most prolific and sophisticated hacking groups, compromising high-profile targets such as SolarWinds.

Feb 24, 2024NewsroomActive Directory / Data Protection

Microsoft has expanded free logging capabilities to all U.S. federal agencies using Microsoft Purview Audit irrespective of the license tier, more than six months after a China-linked cyber espionage campaign targeting two dozen organizations came to light.

“Microsoft will automatically enable the logs in customer accounts and increase the default log retention period from 90 days to 180 days,” the U.S. Cybersecurity and Infrastructure Security Agency (CISA) said.

“Also, this data will provide new telemetry to help more federal agencies meet logging requirements mandated by [Office of Management and Budget] Memorandum M-21-31.”

Microsoft, in July 2023, disclosed that a China-based nation-state activity group known as Storm-0558 gained unauthorized access to approximately 25 entities in the U.S. and Europe as well as a small number of related individual consumer accounts.

“Storm-0558 operates with a high degree of technical tradecraft and operational security,” the company noted. “The actors are keenly aware of the target’s environment, logging policies, authentication requirements, policies, and procedures.”

The campaign is believed to have commenced in May 2023, but detected only a month later after a U.S. federal agency, later revealed to be the State Department, uncovered suspicious activity in unclassified Microsoft 365 audit logs and reported it to Microsoft.

The breach was detected by leveraging enhanced logging in Microsoft Purview Audit, specifically using the MailItemsAccessed mailbox-auditing action that’s typically available for Premium subscribers.

The Windows maker subsequently acknowledged that a validation error in its source code allowed for Azure Active Directory (Azure AD) tokens to be forged by Storm-0558 using a Microsoft account (MSA) consumer signing key, and then use them to penetrate the mailboxes.

The attackers are estimated to have stolen at least 60,000 unclassified emails from Outlook accounts belonging to State Department officials stationed in East Asia, the Pacific, and Europe, Reuters reported in September 2023. Beijing has denied the allegations.

It also faced intense scrutiny for withholding basic-yet-crucial logging capabilities to entities that are on the more expensive E5 or G5 plan, prompting the company to make changes.

“We recognize the vital importance that advanced logging plays in enabling federal agencies to detect, respond to, and prevent even the most sophisticated cyberattacks from well-resourced, state-sponsored actors,” Microsoft’s Candice Ling said. “For this reason, we have been collaborating across the federal government to provide access to advanced audit logs.”

Feb 23, 2024NewsroomRed Teaming / Artificial Intelligence

Microsoft has released an open access automation framework called PyRIT (short for Python Risk Identification Tool) to proactively identify risks in generative artificial intelligence (AI) systems.

The red teaming tool is designed to “enable every organization across the globe to innovate responsibly with the latest artificial intelligence advances,” Ram Shankar Siva Kumar, AI red team lead at Microsoft, said.

The company said PyRIT could be used to assess the robustness of large language model (LLM) endpoints against different harm categories such as fabrication (e.g., hallucination), misuse (e.g., bias), and prohibited content (e.g., harassment).

It can also be used to identify security harms ranging from malware generation to jailbreaking, as well as privacy harms like identity theft.

PyRIT comes with five interfaces: target, datasets, scoring engine, the ability to support multiple attack strategies, and incorporating a memory component that can either take the form of JSON or a database to store the intermediate input and output interactions.

The scoring engine also offers two different options for scoring the outputs from the target AI system, allowing red teamers to use a classical machine learning classifier or leverage an LLM endpoint for self-evaluation.

“The goal is to allow researchers to have a baseline of how well their model and entire inference pipeline is doing against different harm categories and to be able to compare that baseline to future iterations of their model,” Microsoft said.

“This allows them to have empirical data on how well their model is doing today, and detect any degradation of performance based on future improvements.”

That said, the tech giant is careful to emphasize that PyRIT is not a replacement for manual red teaming of generative AI systems and that it complements a red team’s existing domain expertise.

In other words, the tool is meant to highlight the risk “hot spots” by generating prompts that could be used to evaluate the AI system and flag areas that require further investigation.

Microsoft further acknowledged that red teaming generative AI systems requires probing for both security and responsible AI risks simultaneously and that the exercise is more probabilistic while also pointing out the wide differences in generative AI system architectures.

“Manual probing, though time-consuming, is often needed for identifying potential blind spots,” Siva Kumar said. “Automation is needed for scaling but is not a replacement for manual probing.”

The development comes as Protect AI disclosed multiple critical vulnerabilities in popular AI supply chain platforms such as ClearML, Hugging Face, MLflow, and Triton Inference Server that could result in arbitrary code execution and disclosure of sensitive information.