June 2023 Edition

Welcome 🚀

Hello, Cybernauts!

Welcome back to another edition of Cyber, (Mis)Adventures. This is where we dive deeply into cybersecurity, exploring the latest threats, feats, mishaps, and triumphs in this ever-evolving landscape.

️(Mis)Adventures of The Month 🕵️‍♂

This section highlights a particularly interesting or significant cyber (mis)adventure - a zero-day vulnerability in MOVEit transfer exploited to Data theft that had everyone on edge. SMS traffic Pumping Fraud.

1. Zero-Day Vulnerability in MOVEit Transfer Platform

Summary

• 🚨 Mandiant detected exploitation of a zero-day vulnerability in the MOVEit Transfer secure managed file transfer software, leading to data theft.
• 📅 The vulnerability, CVE-2023-34362, was announced by Progress Software Corporation on May 31, 2023.
• 💻 The earliest evidence of exploitation was found on May 27, 2023, resulting in the deployment of web shells and subsequent data theft.
• ⏱ In some instances, data theft occurred within minutes of deploying web shells.
• 🎯 The activity is currently attributed to UNC4857, a newly formed threat cluster with unknown motivations, impacting organizations in Canada, India, and the U.S.
• 🦠 Following the vulnerability exploitation, threat actors have been deploying a newly discovered LEMURLOOT web shell.
• 📊 LEMURLOOT is a web shell written in C# that provides functionality tailored for the MOVEit Transfer software, including generating commands to enumerate files and folders, retrieving configuration information, and creating or deleting a user.
• 🕵️‍♀️ LEMURLOOT is believed to be used to steal data previously uploaded by the users of individual MOVEit Transfer systems.
• ☁️ LEMURLOOT can also steal Azure Storage Blob information, including credentials, from the MOVEit Transfer application settings.
• 🛡️ Mandiant is aware of multiple cases where large volumes of files have been stolen from victims' MOVEit transfer systems and have released a detailed MOVEit Containment and Hardening guide.

Why it matters?

• 🌐 The breach affects many industries globally, suggesting a significant threat to data security.
• 🗂 The potential for immediate data theft after deploying web shells underlines the gravity of the security flaw.
• 🔐 The fact that LEMURLOOT can steal Azure Storage Blob information means potential risks to organizations using Azure for storage.
• 📈 The use of LEMURLOOT for data theft could indicate a new trend in cyber-attack strategies.

Potential Implications

• 💰 Victim organizations could receive ransom emails in the coming days or weeks as the attack is consistent with extortion activities.
• 📜 The data breach might result in the leakage of sensitive information, damaging the reputation of the impacted organizations and possibly leading to financial losses.
• 💼 For the cybersecurity industry, this event underscores the importance of continued vigilance, threat detection, and response capabilities.
• 🛠 Companies using MOVEit Transfer software must implement containment and hardening measures as Mandiant recommends.

2. A Confession Exposes India’s Secret Hacking Industry

Summary

• 🔍 Geneva-based private investigator Jonas Rey was hired to investigate a possible hacking incident concerning an Iranian-born American entrepreneur, Farhad Azima. Azima believed that his email account was hacked following his involvement in exposing Iran's sanctions violations.
• 💻 The investigation pointed towards BellTroX, a New Delhi-based company known for running a hacking-for-hire enterprise. This company had previously been implicated in numerous cyberattacks on various individuals and groups.
• 🕵️ Rey's investigations shed light on a more significant hacking-for-hire industry thriving in India. He secured a confession from a participant in such an operation, confirming that Azima's email account had been infiltrated.
• 📝 The confession came from Aditya Jain, a former worker for Indian cybersecurity firm Appin Security, who later worked as a hacker for hire under the name Cyber Defence and Analytics.
• 👨‍⚖️ Jain admitted to hacking Azima's email account, which became key evidence in Azima's legal battle against the Emirate of Ras Al Khaimah, which was also implicated in the hacking.
• 💼 Despite initial fear of retaliation, Jain decided to come forward publicly, admitting his involvement in court filings.
• 📃 Stuart Page, a private investigator who initially denied any hacking activity, also admitted that hacking had occurred and apologized for misleading the court.
• ⚖️ Azima was granted a retrial in London court, scheduled for spring.
• 📰 Reports from the London Sunday Times and the Bureau of Investigative Journalism suggested that Jain and Rey might have deeper ties to the Indian hacking-for-hire business than previously admitted.

Why it matters

• 🌐 The uncovering of this large hacking-for-hire industry based in India has significant implications for cybersecurity worldwide. The case exemplifies the globalized nature of cybercrime and its potential impact on personal, corporate, and even state security.
• 🏛️ The confession and subsequent retrial for Azima represent a landmark event in the pursuit of justice in cybercrime cases.
• 🕵️‍♂️ The work of private investigators like Jonas Rey highlights the crucial role of independent investigations in uncovering and exposing these operations.

Potential implications

• 🚔 This case could increase international scrutiny and pressure India to address its hacking-for-hire industry.
• ⚠️ It could also lead to an increased focus on cybersecurity measures by individuals, corporations, and governments worldwide to prevent similar incidents.
• ⚖️ The legal outcomes of this case may set new precedents for cybercrime prosecution and the pursuit of justice in similar cases.
• 📈 There may be a surge in demand for cybersecurity services and the development of more advanced cybersecurity tools in response to the growing threat of organized cybercrime.

3. SMS Traffic Pumping Fraud

Summary

1. 📱 SMS Traffic Pumping Fraud, also known as Artificially Inflated Traffic, is a type of fraud involving exploiting phone number input fields for receiving one-time passcodes (OTPs) or app download links via SMS. This can lead to inflated traffic and exploitation of your app.
2. 🤝 It occurs in two scenarios: either the Mobile Network Operator (MNO) is involved in the scheme and has a revenue-sharing agreement with the fraudsters, or the MNO is exploited unknowingly by the fraudsters. It is more common with smaller MNOs.
3. 📈 Signs of an SMS Pumping attack include a sudden increase of messages sent to a block of adjacent numbers, usually in remote countries. You might not see a completed verification cycle for OTP use cases for these OTPs.
4. 🌍 To prevent this type of fraud, disable Geo-Permissions for countries you don't intend to send messages to. This can be managed in your Twilio project in the Console on the Messaging Geographic Permissions page.
5. ⏲️ Implement rate limits to control the number of messages sent per a specific timeframe to the same mobile number range or prefix. User, IP, or device identifier can achieve rate limiting.
6. 🤖 Detect and deter bot traffic with libraries like botd or CAPTCHAs. Make small changes in your user experience to prevent automated scripts and bots, such as confirming email addresses before enrolling in 2FA.
7. ⏱️ Another prevention method is implementing exponential delays between verification retry requests to the same phone number to avoid rapid sending.
8. 📞 Use Carrier Lookup to determine the line type of a number and only send SMS to mobile numbers. You can also use this tool to block carriers causing inflated traffic.
9. 📊 Monitor OTP conversion rates and create alerts. If the verification conversion rate starts to drop, especially in an unexpected country, an alert should be triggered for review.
10. 🛡️ Use Twilio Verify to validate users with SMS, voice, email, push, WhatsApp, and time-based one-time passwords. It can help fight fraud, protect user accounts, and build customer trust. With its Fraud Guard feature, this tool can also help prevent SMS Traffic Pumping Fraud.

Why it Matters

1. Security Risk: SMS Traffic Pumping Fraud represents a significant security risk. Fraudsters exploiting phone number input fields for OTPs or app download links can potentially gain unauthorized access to sensitive information.
2. Financial Impact: The fraudulent activity can lead to inflated traffic, resulting in increased costs for the organization. The organization could also lose revenue if the Mobile Network Operator (MNO) is involved in the scheme.
3. Reputation Risk: If customers are affected by this type of fraud, it could harm the organization's reputation. Customers may lose trust in the organization's ability to protect their information, leading to a loss of business.

Potential Implications

1. Increased Security Measures: Organizations may need to invest in advanced security measures to detect and prevent SMS Traffic Pumping Fraud. This could include implementing rate limits, using tools like botd or CAPTCHAs to deter bot traffic, and using tools like Twilio Verify and Fraud Guard.
2. Operational Changes: Organizations may need to make changes to their operations, such as disabling Geo-Permissions for countries they don't intend to send messages to, implementing exponential delays between verification retry requests, and only sending SMS to mobile numbers.
3. Monitoring and Alerts: Organizations must monitor OTP conversion rates and create alerts for unexpected drops in these rates. This will require resources and may necessitate the development of new monitoring systems or modifying existing ones.
4. Customer Outreach: If customers are affected by the fraud, the organization will need to reach out to these customers, potentially offering compensation or other remedies. This can be significant, particularly for larger organizations with many customers.

LLM's Spotlight 💡 🧠

Industry experts answer your cybersecurity questions.

1. OWASP Top 10 List for Large Language Models

Summary

1. 💉 LLM01:2023 - Prompt Injections: Malicious prompts can bypass filters or manipulate Large Language Models (LLMs), causing them to ignore instructions or perform unintended actions.
2. 🗄️ LLM02:2023 - Data Leakage: LLMs may unintentionally reveal sensitive information, proprietary algorithms, or confidential details in their responses.
3. 📦 LLM03:2023 - Inadequate Sandboxing: A failure to correctly isolate LLMs, especially when they have access to sensitive systems or external resources, opens up the potential for exploitation and unauthorized access.
4. 🛡️ LLM04:2023 - Code Execution Vulnerabilities: LLMs can be exploited to execute malicious code, commands, or actions on the underlying system through natural language prompts.
5. 🕸️ LLM05:2023 - SSRF Vulnerabilities: Unintended requests or access to restricted resources such as internal services, APIs, or data stores can be triggered by exploiting LLMs.
6. 📚 LLM06:2023 - Overreliance on LLM-generated Content: Excessive dependence on LLM-generated content without human oversight can lead to harmful consequences.
7. 🎯 LLM07:2023 - Inadequate AI Alignment: Failing to align the LLM's objectives and behavior with the intended use case can lead to undesired consequences or vulnerabilities.
8. 🔒 LLM08:2023 - Insufficient Access Controls: Inadequate access controls or authentication allows unauthorized users to interact with the LLM, potentially exploiting vulnerabilities.
9. ❌ LLM09:2023 - Improper Error Handling: Exposing error messages or debugging information can reveal sensitive information, system details, or potential attack vectors.
10. 💣 LLM10:2023 - Training Data Poisoning: Malicious manipulation of training data or fine-tuning procedures can introduce vulnerabilities or backdoors into the LLM.

2. The AI Attack Surface Map v1.0

Summary

• 🤖 This resource provides a framework for conceptualizing attacks on A.I. systems, especially those using recent technologies like Langchain and ChatGPT.
• ⏳ It acknowledges that A.I. system security is still in its early stages, with technologies such as ChatGPT only being released recently.
• 🔧 The A.I. attack surface comprises multiple components, including A.I. Assistants, Agents, Tools, Models, and Storage, each having unique vulnerabilities.
• 🗣️ The primary method of attack for A.I. systems is through natural language, marking a new class of vulnerabilities.
• 🕵️‍♂️ A.I. Assistants, who manage individuals' digital lives, pose significant privacy and security risks due to the sensitive data they access and manage.
• 💻 Agents, or A.I. entities with specific purposes, are susceptible to attacks that could make them perform unintended actions.
• 🛠️ Tools, the Agents' capabilities, can be misused through prompt injections to perform unintended tasks.
• 🎯 Attacking models, or manipulating A.I. to behave negatively, is a mature practice in the A.I. security space.
• 💽 Storage mechanisms such as Vector Databases, which can't fit everything into models, also pose security risks.
• 🎯 Specific types of attacks include prompt injection, training attacks, altering agent routing, executing arbitrary commands, attacking embedding databases, and others.

Why it matters

• 🚀 A.I. systems are integrating rapidly into society, necessitating understanding and preparedness from a security standpoint.
• 💡 It's crucial for CISOs to be aware of the potential attack surfaces within an A.I. system and ensure they have appropriate defenses in place.
• 🌐 A.I. Assistants, who have access to sensitive data, pose considerable risks if compromised, which could have significant consequences for the privacy and security of an organization.
• ⚠️ Security risks tied to A.I. systems, such as Agents and Tools, could impact business operations and integrity if not appropriately managed.

Potential implications

• 📈 As A.I. continues to grow in prominence, the volume of security challenges for organizations will likely increase.
• 🏦 CISOs will have to allocate resources for understanding, assessing, and defending against potential AI-related security risks.
• 🔒 Organizations could face serious privacy breaches and operational disruptions if A.I. systems are not secured effectively.
• 🤝 There may be increased collaboration between academia and organizations to mitigate model-based attacks.
• 💼 The rise in A.I. integration may drive the demand for professionals with specialized skills in A.I. security, affecting hiring strategies in organizations.

3. Can LLMs B.E. Attack?

Summary

• 🤖 The video discusses potential security risks associated with large language models (LLMs) like Google Bard and GPT, hosted on public cloud service providers like AWS.
• 🗣️ LLMs are essentially giant databases that interpret and respond to user prompts, making them potential targets for various types of attacks.
• ⚠️ The first type of attack is by manipulating input prompts, possibly leading to unauthorized behaviors or code execution.
• 📈 The second type of attack is data-based, including data leakage, data poisoning, and trained data leakage, which could expose user information or corrupt the training data.
• 🔐 The third category is attacks on the LLM application itself, such as vulnerabilities introduced by developers, open-source library vulnerabilities, and identity and access management issues.
• ☁️ The fourth category is infrastructure attacks on the hosting platform, such as public cloud service providers.

Why it matters?

• 🌐 Large Language Models are integral parts of many digital services and systems; thus, their security is crucial to prevent misuse and protect user data.
• 🕵️ The potential for manipulating or exploiting these models could severely affect user privacy and data integrity.
• 🛠️ Developers and organizations employing LLMs must be aware of these risks to ensure they implement necessary protections.

Potential implications

• 💻 Given the ubiquity of A.I. systems in various sectors, a successful attack could lead to severe consequences, including exposure of sensitive information, disruption of services, or manipulation of A.I. responses.
• 📚 If unchecked, these vulnerabilities could erode public trust in A.I. and machine learning technologies, impeding their adoption and usefulness.
• 🛡️ Addressing these vulnerabilities will require a multi-faceted approach, combining enhanced security measures, rigorous testing, and continuous monitoring of A.I. systems.

Closing Notes 💌

We hope you found this month's edition of our newsletter informative and valuable.

If you have any feedback, suggestions, or topics that you would like us to cover in our future newsletters, please don't hesitate to reach out to me.

Also, don't forget to share this newsletter with your colleagues, friends, and family members who might benefit from the information we share.

Thank you for your continued support, and stay safe!

Social Media Links 👥

Keep the conversation going. Let's connect!
Twitter: https://twitter.com/IshanGirdhar
LinkedIn: https://www.linkedin.com/in/ishangirdhar/

Stay safe in cyberspace, and see you next month!

Amor Fati
Ishan Girdhar