Global Security Weekly: $290M DeFi Attack and Supply Chain Crisis

Introduction: Old Vulnerabilities Persist as New Threats Emerge

As you review this week's security incident reports, an unsettling sense of déjà vu sets in — many attack techniques look strikingly familiar, as if they should have been definitively patched years ago, yet they remain effective with only minor modifications. The same vulnerabilities, the same mistakes, exploited again and again across different contexts.

This week's ThreatsDay security bulletin covers more than 25 new threat stories, with three core incidents drawing particular attention: a massive $290 million DeFi protocol hack, a "Living off the Land" (LotL) abuse attack targeting macOS, and the large-scale exposure of the ProxySmart SIM farm. Together, these events point to a harsh reality — the cybersecurity balance of power is tilting toward attackers, and the involvement of AI technology is making this trend even more complex.

Core Incident One: $290M DeFi Protocol Suffers Epic Attack

The most high-profile event this week was undoubtedly a DeFi (Decentralized Finance) protocol attack involving $290 million. The attacker exploited known logic flaws in smart contracts, draining the protocol's liquidity pools through carefully crafted transaction sequences in an extremely short timeframe.

Notably, the vulnerability pattern exploited in this attack was not a new discovery. Security researchers pointed out that similar reentrancy attacks and oracle manipulation techniques have been repeatedly discussed and warned about over the past several years, yet many DeFi projects still fail to include these known risks in their audit scope during rapid iteration cycles. The attack methods were simple yet still effective — precisely the most alarming issue in the current Web3 security landscape.

Even more concerning, some security analysts have discovered that the attacker may have used AI tools to automate the scanning of smart contract vulnerabilities, significantly lowering the barrier to entry for attacks. When AI is deployed on the offensive side, traditional manual code audit models face unprecedented efficiency challenges.

Core Incident Two: macOS LotL Attack and ProxySmart SIM Farm Exposure

The macOS platform was reported this week to have suffered large-scale "Living off the Land" attacks. This attack method does not rely on external malware but instead abuses legitimate tools built into the operating system — such as script interpreters and system management commands — to carry out malicious operations. Since all operations are completed through the system's "native" tools, traditional antivirus software and endpoint protection solutions find it extremely difficult to identify and intercept them.

Attackers leveraged macOS built-in tool chains including osascript, curl, and python3 to achieve a complete attack chain from data exfiltration to persistent access. Security experts warned that as macOS continues to grow its market share in enterprise environments, LotL attacks targeting the platform will become a significant threat trend in 2025.

Meanwhile, the exposure of the ProxySmart SIM farm revealed another disturbing piece of attack infrastructure. SIM farms build proxy networks using large numbers of physical SIM cards, providing attackers with communication channels that are difficult to trace, widely used in SMS phishing, verification code interception, fake account registration, and other malicious activities. ProxySmart was found to operate large-scale SIM card pools across multiple regions worldwide, providing critical anonymization infrastructure for the cybercrime ecosystem.

Deep Analysis: Supply Chain Security Has Become the Weakest Link

Looking across the 25+ security incidents in this week's bulletin, a recurring theme is the proliferation of supply chain attacks. As security researchers summarized: "The supply chain is in chaos. Those packages you never reviewed are stealing data, planting backdoors, and continuing to spread. Attacking the systems behind applications is far easier than breaching the applications themselves."

This assessment precisely captures the core contradiction of the current security landscape. Modern software development is heavily dependent on open-source components and third-party libraries; a typical enterprise application may contain hundreds of direct dependencies and thousands of indirect dependencies. Attackers have realized that rather than expending significant effort to find zero-day vulnerabilities in target applications, it is far more effective to poison upstream package repositories — this "indirect attack" strategy is lower in cost and broader in impact.

Multiple recent malicious package incidents on npm and PyPI have confirmed this trend. Attackers register malicious packages with names highly similar to well-known packages (typosquatting) or directly hijack maintainer accounts to tamper with the contents of legitimate packages. Victims unknowingly introduce malicious code into their own projects, leading to data breaches and backdoor implantation.

AI plays a double-edged sword role in this domain as well. On one hand, code generation tools based on large language models may inadvertently recommend dependencies containing vulnerabilities; on the other hand, AI-driven security scanning tools are also helping developers identify suspicious dependencies more quickly. This AI-empowered offensive-defensive contest is redefining the boundaries of software supply chain security.

Technology Trends: AI Accelerates Offensive-Defensive Asymmetry

From this week's threat landscape, AI technology is reshaping the cybersecurity landscape in multiple ways:

Offensive AI applications are becoming increasingly mature. Automated vulnerability discovery, intelligent phishing email generation, and deepfake identity verification bypass technologies have moved from proof-of-concept to real-world deployment. Behind the $290 million DeFi attack, AI-assisted contract analysis may have played a critical role.

Defensive AI bottlenecks persist. Although AI shows enormous potential in threat detection and anomalous behavior analysis, the success of LotL attacks demonstrates that when malicious behavior and legitimate operations overlap significantly at the feature level, the false positive and false negative rates of AI models remain urgent problems to solve.

AI governance for supply chain security has become a new focal point. How to leverage AI for real-time reputation assessment, behavioral analysis, and risk scoring of massive open-source components is becoming a new competitive arena for security vendors.

Outlook: Building Defense-in-Depth for the AI Era

This week's security bulletin sounds the alarm once again: cybersecurity is not a problem that can be solved "once and for all." Old vulnerabilities resurface in new contexts, and simple attack methods remain effective — this indicates that the industry still has enormous room for improvement in fundamental security practices.

Looking ahead, several directions deserve focused attention. First, the DeFi space urgently needs stricter smart contract audit standards and real-time monitoring mechanisms, with AI-assisted formal verification tools becoming essential. Second, endpoint security must comprehensively shift from "signature-based" detection models to "behavior-based" analysis models to address the challenge of fileless attacks such as LotL. Finally, software supply chain security requires industry-wide collaboration to establish full-chain trust mechanisms from source code to deployment.

When attackers begin systematically leveraging AI to boost efficiency, defenders must embrace AI even more urgently to bridge the security capability gap. In this endless offensive-defensive contest, only continuous evolution can prevent becoming the next "old story" in a security bulletin.