新闻

Microsoft: Hackers abusing AI at every stage of cyberattacks

· BleepingComputer

Microsoft's latest threat intelligence research reveals a concerning escalation in how cybercriminals are weaponizing artificial intelligence across every phase of their attack operations. The technology giant's security team has documented widespread adoption of generative AI tools by threat actors to streamline malicious activities, reduce technical barriers, and accelerate the pace of cyberattacks. The comprehensive report demonstrates that attackers are leveraging AI language models for diverse criminal purposes, including reconnaissance operations, crafting sophisticated phishing campaigns, developing malicious infrastructure, creating malware, and conducting post-breach activities. These AI-powered capabilities enable criminals to draft convincing phishing messages, translate content for global operations, analyze stolen data efficiently, debug malicious code, and automate infrastructure configuration tasks. Particularly notable are the activities of North Korean threat groups Jasper Sleet and Coral Sleet, who have integrated AI into elaborate remote IT worker infiltration schemes. These operations demonstrate sophisticated use of generative AI to create realistic professional identities, craft culturally appropriate names and email formats, and generate convincing resumes tailored to specific job requirements. The groups use AI to analyze job postings and extract required skills, then customize fake personas to match those specifications perfectly. The malware development landscape is also being transformed by AI integration. Cybercriminals are employing AI coding assistants to generate and refine malicious software, troubleshoot programming errors, and port malware components between different programming languages. Some experimental malware samples show evidence of AI-enabled dynamic script generation and runtime behavior modification, representing a significant evolution in threat sophistication. When AI safety mechanisms attempt to prevent malicious use, attackers are developing jailbreaking techniques to circumvent these protections and trick language models into generating harmful content. Microsoft researchers have also observed early experimentation with agentic AI systems that can perform autonomous tasks and adapt to changing conditions, though current implementations focus primarily on decision-making rather than fully autonomous attacks. The implications extend beyond traditional cybersecurity concerns. Since many AI-powered campaigns exploit legitimate access credentials, Microsoft recommends treating these activities as insider threats requiring specialized detection and response strategies. Organizations need to focus on identifying abnormal credential usage patterns, strengthening identity systems against phishing attacks, and securing their own AI infrastructure from potential targeting. This trend is not isolated to Microsoft's observations. Google has reported similar abuse of their Gemini AI platform across attack lifecycles, while Amazon documented cases where threat actors used multiple generative AI services in campaigns that successfully compromised over 600 FortiGate firewalls. The convergence of these reports from major technology companies indicates that AI-powered cybercrime has reached a critical inflection point, requiring immediate attention from security professionals and policymakers alike.

Microsoft's cybersecurity researchers have published alarming findings about the rapid adoption of artificial intelligence by cybercriminals, revealing how threat actors are systematically integrating AI tools across every stage of their attack operations. The comprehensive threat intelligence report documents a fundamental shift in the cybercrime landscape, where AI serves as a force multiplier that dramatically reduces technical barriers and accelerates malicious activities.

The research identifies widespread use of generative AI language models for producing text, code, and media content in support of criminal operations. Attackers are leveraging these capabilities for reconnaissance activities, crafting sophisticated phishing campaigns, developing attack infrastructure, creating and debugging malware, and conducting post-compromise data analysis. This represents a significant evolution from traditional manual methods to AI-augmented criminal enterprises.

North Korean threat groups have emerged as particularly sophisticated adopters of AI technology. Jasper Sleet (Storm-0287) and Coral Sleet (Storm-1877) have integrated generative AI into elaborate remote IT worker infiltration schemes designed to gain employment at Western companies. These operations demonstrate remarkable sophistication in using AI to generate realistic professional identities, including culturally appropriate names, email formats, and detailed resumes tailored to specific job requirements.

The groups employ strategic prompting techniques to extract maximum value from AI platforms. They generate comprehensive lists of culturally specific names and corresponding email address formats to create convincing professional personas. Additionally, they use AI to analyze job postings for software development and IT roles, extracting required skills and qualifications to customize their fake identities accordingly.

Malware development has been significantly transformed by AI integration. Cybercriminals are using AI coding assistants to generate malicious software, refine existing code, troubleshoot programming errors, and port malware components between different programming languages. Some experimental samples show evidence of AI-enabled dynamic behavior, including runtime script generation and adaptive modification capabilities that represent a new frontier in threat sophistication.

Coral Sleet has demonstrated particular innovation in using AI for infrastructure development, rapidly generating convincing fake company websites, provisioning attack infrastructure, and testing deployment configurations. This automation significantly reduces the time and technical expertise required to establish credible criminal operations.

When AI safety mechanisms attempt to prevent malicious use, attackers have developed sophisticated jailbreaking techniques to circumvent these protections. These methods trick language models into generating harmful content despite built-in safeguards, highlighting the ongoing arms race between AI safety measures and criminal innovation.

Microsoft researchers have also observed early experimentation with agentic AI systems capable of performing autonomous tasks and adapting to changing conditions. While current implementations focus primarily on decision-making rather than fully autonomous attacks, this represents a concerning trend toward more sophisticated AI-powered criminal capabilities.

The implications extend far beyond traditional cybersecurity concerns. Since many AI-powered campaigns exploit legitimate access credentials and mimic normal user behavior, Microsoft recommends treating these activities as insider threats requiring specialized detection and response strategies. Organizations must focus on identifying abnormal credential usage patterns, strengthening identity systems against sophisticated phishing attacks, and securing their own AI infrastructure from potential targeting.

This trend is not isolated to Microsoft's observations. Google has reported similar abuse of their Gemini AI platform across attack lifecycles, while Amazon documented cases where threat actors used multiple generative AI services in campaigns that successfully compromised over 600 FortiGate firewalls. The convergence of these reports from major technology companies indicates that AI-powered cybercrime has reached a critical inflection point.

The cybersecurity industry must rapidly adapt to address these evolving threats. Traditional security measures designed for human-operated attacks may prove insufficient against AI-augmented criminal operations that can operate at unprecedented scale and sophistication. Organizations need to invest in advanced detection capabilities, enhance their security awareness training to address AI-generated threats, and develop new defensive strategies specifically designed to counter AI-powered attacks.

参考链接