This Week in Computing: Malware Gone Wild
Malware is sneaky AF. It tries to hide itself and cover up its actions. It detects when it is being studied in a virtual sandbox, and so it sits still to evade detection. But when it senses a less secure environment — such as an unpatched Windows 7 box — it goes wild, as if possessing a split personality.
In other words, malware can no longer be fully understood simply by studying it in a lab setting, asserted the University of Maryland Associate professor Tudor Dumitras, in a recently posted talk from USENIX‘s last-ever Enigma security and privacy conference.
Today, most malware is examined by examining execution traces that the malicious program generates (“Dynamic Malware Analysis”). This is usually done in a controlled environment, such as a sandbox or virtual machine. Such analysis creates the signatures to describe the behavior of the malicious software.
The malware community, of course, has been long hip to this scrutiny, and has developed an evasion technique known as red pills, which helps malware detect when it is in a controlled environment, and change its behavior accordingly.
As a result, many of the signatures used for commercial malware detection packages may not be able to adequately to identify malware in all circumstances, depending on what traces the signature actually captured.
What we really need, Dumitras said, is execution traces from the wild. Dumitras led a study that collected info on real-world attacks, consisting of over 7.6 million traces from 5.4 million users.
“Sandbox traces can not account for the range of behaviors encountered in the wild.”
They had found that, as Dumitras expected, traces collected in a sandbox rarely capture the full behavior of malware in the wild.
In the case of Wannacry ransom attack, for instance, sandbox tracing only caught 18% of all the actions that the randomware attack executed in the wild.
For the keepers of malware detection engines, Dumitras advised using traces from multiple executions in the wild. He advised using three separate traces, as diminishing returns set in after that.
Full video of the talk here:
“So far, having an AI CEO hasn’t had any catastrophic consequences for NetDragon Websoft. In fact, since Yu’s appointment, the company has outperformed Hong Kong’s stock market.” — The Hustle, on replacing CEOs with AI Chatbots.
AI “Latent space embeddings end up being a double-edged sword. They allow the model to efficiently encode and use a large amount of data, but they also cause possible problems where the AI will spit out related but wrong information.” — Geek Culture, on why ChatGPT lies.
“We think someone who writes for a living needs to constantly be thinking about the best way to express complex ideas in their own words.” – Wired, on its editorial use of generative AI.
AI Hype holes?
The biggest one is that building the model underneath GPT-4 takes 18 months. So it can’t tell you anything new. Can’t help you on your fantasy football league. Can’t tell you what happened in the stock market today.
That won’t get fixed soon, I hear from AI… https://t.co/XPKUYitNnc
— Robert Scoble (@Scobleizer) March 17, 2023
“I think with Kubernetes, we did a decent job on the backend. But we did not get developers, not one little bit. That was a missed opportunity to really bring the worlds together in a natural way” — Kubernetes co-founder Craig McLuckie, on how the operations-centric Kubernetes perplexed developers (See: YAML), speaking at a Docker press roundtable this week.
McLuckie also noted that 60% of machine learning workloads now run on Kubernetes.
“After listening to feedback and consulting our community, it’s clear that we made the wrong decision in sunsetting our Free Team plan. Last week we felt our communications were terrible but our policy was sound. It’s now clear that both the communications and the policy were wrong, so we’re reversing course and no longer sunsetting the Free Team plan” —Docker, responding to the outcry in the open source community over the suspension of its free Docker Hub tier for teams.
“Decorators are by far the biggest new feature, making it possible to decorate classes and their members to make them more easily reusable. […] Decorators are just syntactic glue aiming to simplify the definition of higher-order functions” — Software Engineer Sergio De Simone on the release of TypeScript 5.0, in InfoQ.
— programmerjokesofficial (@programmerjoke9) March 19, 2023
“If these details cannot be hidden from you, and you need to build a large knowledge base around stuff that does not directly contribute to implementing your program, then choose another platform.” — Hacker News commenter, on the needless complexity that came with using Microsoft Foundation Classes (MFC) for C++ coding.
Ken Thompson “has a long and storied history of trolling the computer industry […] he revealed, during his Turing Award lecture, that he had planted an essentially untraceable back door in the original C compiler… and it was still there.” — Liam Proven, The Register.