Analysis Summary

A vulnerability in Microsoft 365 Copilot that has now been patched has come to light. This vulnerability could allow sensitive user data to be stolen through the use of an ASCII smuggling technique.

A new method called ASCII Smuggling makes use of unique Unicode characters that mimic ASCII but aren't displayed in user interfaces. Researchers said that this implies a threat actor may utilize a large language model to generate data that is invisible to the user and insert it into hyperlinks that can be clicked. In essence, this method stages the data for exfiltration.

The complete attack combines several attack techniques to create a trustworthy chain of exploits. This entails taking the subsequent actions:

• Use malicious material hidden in a document given over the conversation to initiate prompt injection.
• Using a prompt injection payload to tell Copilot to go through further documents and emails.
• Using ASCII smuggling to trick a user into clicking on a link that will allow the theft of important information to a third-party server.

The attack's ultimate result is the possibility of sending private information—such as multi-factor authentication (MFA) codes—to a server under the control of the adversary. Since then, Microsoft has dealt with the problems after making a responsible disclosure in January 2024. The revelation coincides with the demonstration of proof-of-concept (PoC) assaults against Microsoft's Copilot system, which highlights the necessity of continuously monitoring dangers associated with artificial intelligence (AI) technologies to influence answers, exfiltrate private data, and evade security measures.

The techniques enable malevolent actors to carry out indirect prompt injection and retrieval-augmented generation (RAG) poisoning, which can result in remote code execution attacks that can take complete control of Microsoft Copilot and other AI apps. In a best-case scenario, Copilot might be tricked into sending visitors to phishing pages by an outside hacker with code execution skills. Making the AI become a spear-phishing machine is perhaps one of the most inventive attacks. Using a red-teaming technique known as LOLCopilot, an attacker can send phishing messages that imitate the style of a compromised user if they have access to the victim's email account.

Microsoft has also said that provided threat actors have prior knowledge of the Copilot name or URL, publicly accessible Copilot bots made with Microsoft Copilot Studio and without any authentication safeguards may provide a means for them to obtain sensitive data. To avoid data leaks from Copilots (previously Power Virtual Agents), enterprises should assess their exposure to risk and risk tolerance. Then, they should enable Data Loss Prevention and other security controls following their assessment to govern the creation and publication of Copilots.

Impact

• Data Exfiltration
• Security Bypass
• Code Execution
• Sensitive Data Theft

Affected Vendors

Remediation

• Organizations must test their assets and apply the available security patches or mitigation steps as soon as possible.
• Implement multi-factor authentication to add an extra layer of security to login processes.
• Regularly monitor network activity for any unusual behavior, as this may indicate that a cyberattack is underway.
• Organizations must stay vigilant and follow best practices for cybersecurity to protect their systems and data from potential threats. This includes regularly updating software and implementing strong access controls and monitoring tools.
• Develop a comprehensive incident response plan to respond effectively in case of a security breach or data leakage.
• Maintain regular backups of critical data and systems to ensure data recovery in case of a security incident.
• Adhere to security best practices, including the principle of least privilege, and ensure that users and applications have only the necessary permissions.
• Establish a robust patch management process to ensure that security patches are evaluated, tested, and applied promptly.
• Conduct security audits and assessments to evaluate the overall security posture of your systems and networks.
• Implement network segmentation to contain and isolate potential threats to limit their impact on critical systems.