[cb22] 「協調���れた脆弱性開示の現在と未来」国際的なパネルディスカッション（2）by Allan Friedman

2021年10月、Lazarusグループに関連する可能性が高いユニークなローダーであるWSLinkの最初の分析を公開。ほとんどのサンプルは難読化され、高度な仮想マシン（VM）難読化機能で保護されている。サンプルには明確なアーティファクトが含まれておらず、当初は難読化を公的に知られているVMと関連付けなかったが、後にそれをCodevirtualizerに接続することに成功。このVMは、ジャンクコードの挿入、仮想オペランドの暗号化、仮想オペコードの重複、難読化手法仮想命令のマージ、ネストされたVMなど、いくつかの追加の難読化技術を導入する。 本発表では、VMの内部を分析し、合理的な時間で難読化技術を「見抜く」ための半自動化されたアプローチについて説明する。また、難読化されたバイトコードと難読化されていないバイトコードを比較し、本手法の有効性を紹介する。われわれの手法は、仮想オペコードのセマンティクスを抽出する既知の難読化解除手法に基づいており、単純化規則によるシンボリック実行を使用。さらに、バイトコードチャンクとVMの内部構成を記号ではなく、具体的な値として扱い、既知の難読化手法で追加の難読化技術を自動的に処理できるようにする。

In October 2021, we published the first analysis of Wslink – a unique loader likely linked to the Lazarus group. Most samples are packed and protected with an advanced virtual machine (VM) obfuscator; the samples contain no clear artifacts and we initially did not associate the obfuscation with a publicly known VM, but we later managed to connect it to CodeVirtualizer. This VM introduces several additional obfuscation techniques such as insertion of junk code, encoding of virtual operands, duplication of virtual opcodes, opaque predicates, merging of virtual instructions, and a nested VM. Our presentation analyzes the internals of the VM and describes our semi automated approach to “see through” the obfuscation techniques in reasonable time. We demonstrate the approach on some bytecode from a protected sample and compare the results with a non-obfuscated sample, found subsequent to starting our analysis, confirming the method’s validity. Our solution is based on a known deobfuscation method that extracts the semantics of the virtual opcodes, using symbolic execution with simplifying rules. We further treat the bytecode chunks and some internal constructs of the VM as concrete values instead of as symbolic ones, enabling the known deobfuscation method to deal with the additional obfuscation techniques automatically.

Kimsuky is a North Korean APT possibly controlled by North Korea's Reconnaissance General Bureau. Based on reports from the Korea Internet & Security Agency (KISA) and other vendors, TeamT5 identified that Kimsuky's most active group, CloudDragon, built a workflow functioning as a "Credential Factory," collecting and exploiting these massive credentials. The credential factory powers CloudDragon to start its espionage campaigns. CloudDragon's campaigns have aligned with DPRK's interests, targeting the organizations and key figures playing a role in the DPRK relationship. Our database suggested that CloudDragon has possibly infiltrated targets in South Korea, Japan, and the United States. Victims include think tanks, NGOs, media agencies, educational institutes, and many individuals. CloudDragon's "Credential Factory" can be divided into three small cycles, "Daily Cycle," "Campaign Cycle," and "Post-exploit Cycle." The"Daily Cycle" can collect massive credentials and use the stolen credentials to accelerate its APT life cycle. In the "Campaign Cycle," CloudDragon develops many new malware. While we responded to CloudDragon's incidents, we found that the actor still relied on BabyShark malware. CloudDragon once used BabyShark to deploy a new browser extension malware targeting victims' browsers. Moreover, CloudDragon is also developing a shellcode-based malware, Dust. In the "Post-exploit Cycle," the actor relied on hacking tools rather than malicious backdoors. We also identified that the actor used remote desktop software to prevent detection. In this presentation, we will go through some of the most significant operations conducted by CloudDragon, and more importantly, we will provide possible scenarios of future invasions for defense and detection.

Social media is no doubt a critical battlefield for threat actors to launch InfoOps, especially in a critical moment such as wartime or the election season. We have seen Bot-Driven Information Operations (InfoOps, aka influence campaign) have attempted to spread disinformation, incite protests in the physical world, and doxxing against journalists. China's Bots-Driven InfoOps, despite operating on a massive scale, are often considered to have low impact and very little organic engagement. In this talk, we will share our observations on these persistent Bots-Driven InfoOps and dissect their harmful disinformation campaigns circulated in cyberspace. In the past, most bots-driven operations simply parroted narratives of the Chinese propaganda machine, mechanically disseminating the same propaganda and disinformation artifacts made by Chinese state media. However, recently, we saw the newly created bots turn to post artifacts in a livelier manner. They utilized various tactics, including reposting screenshots of forum posts and disguised as members of “Milk Tea Alliance,” to create a false appearance that such content is being echoed across cyberspace. We particularly focus on an ongoing China's bots-driven InfoOps targeting Taiwan, which we dub "Operation ChinaRoot." Starting in mid-2021, the bots have been disseminating manipulated information about Taiwan's local politics and Covid-19 measures. Our further investigation has also identified the linkage between Operation ChinaRoot and other Chinese state-linked networks such as DRAGONBRIDGE and Spamouflage.

Malwares written in Go is increasing every year. Go's cross-platform nature makes it an opportune language for attackers who wish to target multiple platforms. On the other hand, the statically linked libraries make it difficult to distinguish between user functions and libraries, making it difficult for analysts to analyze. This situation has increased the demand for Go malware classification and exploration. In this talk, we will demonstrate the feasibility of computing similarity and classification of Go malware using a newly proposed method called gimpfuzzy. We have implemented "gimpfuzzy", which incorporates Fuzzy Hashing into the existing gimphash method. In this talk, we will verify the discrimination rate of the classification using the proposed method and confirm the validity of the proposed method by discussing some examples from the classified results. We will also discuss issues in Go-malware classification.

Goで書かれたマルウェアは年々増加している。Goはクロスプラットフォームの性質を持っており、複数のプラットフォームを標的にしたい攻撃者にとって好都合な言語である。その一方で、ライブラリが静的にリンクされていることからユーザ関数とライブラリの区別が難しく、アナリストにとって解析が困難である。そうした状況で、Goマルウェアの分類や探索の需要が高まっている。 本講演ではgimpfuzzyという新たな提案手法を用いてGoマルウェアに対し類似性の計算や分類が可能であることを検証する。われわれは既存手法であるgimphashにFuzzy Hashingを組み込んだ「gimpfuzzy」を新たに実装した。講演では提案手法を利用した分類の判別率を検証し、分類された結果の中からいくつかの事例を取り上げその妥当性について確認する。また、Goマルウェアの分類における課題についても検討を行う予定である。

This document discusses the results of long-term scanning and analysis of Winnti 4.0 and ShadowPad malware command and control (C2) protocols. It finds that Winnti 4.0 C2s primarily use TLS, HTTPS, and HTTP, while ShadowPad variants primarily use TCP, HTTPS, and HTTP. Analysis of the protocols reveals encryption methods, packet structures, and server-side functionality. Over time, the number and distribution of active C2s changed, likely in response to research publications and incident response actions. The document advocates for anonymization techniques and merits and risks of future research publications.

We are swamped with new types of malware every day. The goal of malware analysis is not to reveal every single detail of the malware. It is more important to develop tools for efficiency or introduce automation to avoid repeating the same analysis process. Therefore, malware analysts usually actively develop tools and build analysis systems. On the other hand, it costs a lot for such tool developments and system maintenance. Incident trends change daily, and malware keeps evolving. However, it is not easy to keep up with new threats. Malware analysts spend a long time maintaining their analysis systems, and it results in reducing their time for necessary analysis of new types of malware. To solve these problems, we incorporate DevOps practices into malware analysis to reduce the cost of system maintenance by using CI/CD and Serverless. This presentation shares our experience on how CI/CD, Serverless, and other cloud technologies can be used to streamline malware analysis. Specifically, the following case studies are discussed. * Malware C2 Monitoring * Malware Hunting using Cloud * YARA CI/CD system * Malware Analysis System on Cloud * Memory Forensic on Cloud Through the above case studies, we will share the benefits and tips of using the cloud and show how to build a similar system using Infrastructure as Code (IaC). The audience will learn how to improve the efficiency of malware analysis and build a malware analysis system using Cloud infrastructure.

In November 2019, I started monitoring the Bitcoin operation by the adversaries who hid IP addresses of their C&C server in the blockchain. In June 2020, I started collaborating with Professor Christian Doerr of the Hasso Plattner Institute based on the idea of redirecting C&C server communication to a sinkhole server (called takeover), and we successfully achieved this in August. However, the adversaries quickly took evasive action, where they managed to implement an evasion mechanism in only two weeks and restarted their attack. Although we could not conduct our takeover, our monitoring system could worked well. The end of their attack was brought upon by the surge in Bitcoin prices. Due to the fees for the Bitcoin miners, a transaction had reduced the adversaries' profits, and we confirmed the last C&C update was in January 2021 and the abandonment of the attack infrastructure came in March. Since then, no similar attacks have been observed by my monitoring system. Although this attack has already concluded and is unlikely to restart unless the value of Bitcoin declines, I would like to share the know-how I have learned through the direct confrontation with the adversaries. That is, at the time of the confrontation with them, this attack was highly novel, and the adversaries themselves did not fully understand the best solution for its' operation. They needed to evolve their tactics, techniques, and procedures (TTPs) while operating the system. We carefully analyzed their TTPs and tried to catch them off their guard. Even more troublesome was the need to understand as quickly as possible what they intended to do each time they were affected by the Bitcoin halving or making a simple operational error. This presentation is a culmination my insights learned from interactions with these adversaries and I am looking forward to sharing this information with everyone.

本研究では 2019 年 11 月から C&C サーバーの IP アドレスをブロックチェーンに隠ぺいした攻撃者のビットコイン運用監視を開始した。2020 年 6 月に C&C サーバ通信をシンクホールサーバへ直接誘導する (テイクオーバーと呼ぶ) アイデアによる国際協業を Hasso Plattner Institute の Christian Doerr 教授と開始し、8 月にテイクオーバーに成功した。攻撃者のテイクオーバー回避は早く、約 2 週間で回避メカニズムを実装し攻撃を再開した。テイクオーバーは回避されてしまったが、ビットコイン運用監視は機能し続けた。この攻撃の終息はビットコイン高騰がきっかけとなった。ビットコイン取引における採掘者への手数料が利益を圧迫する要因となり、2021 年 1 月に最後の C&C 情報の更新、3 月に攻撃インフラ放棄を確認した。その後、本研究の監視範囲において同種の攻撃は観察されていない。 この攻撃はすでに終息し、ビットコインの価値が下がらない限り再開される可能性は低いが、本講演では攻撃者との直接対峙により得られたノウハウを共有したい。つまり、攻撃者と対峙していた当時、この攻撃は新規性が高く、攻撃者自身も最適な運用方法を理解できていなかった。運用しながら攻撃手法を進化させる必要があり、われわれも攻撃手法を慎重に分析しながら隙を狙っていた。さらに厄介なのが、攻撃者がビットコイン半減期の影響を受けたり、単純な運用ミスをしたりして、そのたびに、われわれも攻撃者の意図を可能な限り早く理解しなければならなかったという点だ。この対峙により得られた知見は、本講演者による CODE BLUE 講演でも活かしており、本質的なノウハウとして共有する。

Smartian is a tool that enhances smart contract fuzzing with static and dynamic data-flow analyses. It integrates static analysis to identify promising sequences of function calls for generating initial fuzzing seeds. It then uses dynamic analysis to mutate function arguments to realize expected data flows across functions. Smartian implements bug oracles for 13 classes of smart contract bugs. Evaluation shows Smartian outperforms other fuzzers and symbolic executors on benchmarks with known bugs, demonstrating the effectiveness of integrating static and dynamic analyses for smart contract fuzzing.

Imagine a world where a security researcher becomes aware of a security vulnerability, impacting thousands of Open Source Software (OSS) projects, and is enabled to both identify and fix them all at once. Now imagine a world where a vulnerability is introduced into your production code and a few moments later you receive an automated pull request to fix it. Hundreds of thousands of human hours are invested every year in finding common security vulnerabilities with relatively simple fixes. These vulnerabilities aren't sexy, cool, or new, we've known about them for years, but they're everywhere! The scale of GitHub and tools like CodeQL (GitHub's code query language) enable one to scan for vulnerabilities across hundreds of thousands of OSS projects, but the challenge is how to scale the triaging, reporting, and fixing. Simply automating the creation of thousands of bug reports by itself isn't useful, and would be even more of a burden on volunteer maintainers of OSS projects. Ideally, the maintainers would be provided with not only information about the vulnerability, but also a fix in the form of an easily actionable pull request. When facing a problem of this scale, what is the most efficient way to leverage researcher knowledge to fix the most vulnerabilities across OSS? This talk will cover a highly scalable solution - automated bulk pull request generation. We'll discuss the practical applications of this technique on real world OSS projects. We'll also cover technologies like CodeQL and OpenRewrite (a style-preserving refactoring tool created at Netflix and now developed by Moderne). Let's not just talk about vulnerabilities, let's actually fix them at scale. This work is sponsored by the new Dan Kaminsky Fellowship; a fellowship created to celebrate Dan's memory and legacy by funding open-source work that makes the world a better (and more secure) place.