Flipping The Script On AI By Using Prompt Engineering Techniques To Have Generative AI Ask You Questions And Get Even Better Answers Out Of The AI

Flipped interaction as a prompt engineering technique that turns generative AI on its head.

Flipping the script.

This well-known catchphrase refers to turning things on their head and doing nearly the opposite of what is normally done. Up becomes down, down becomes up. There can be lots of good reasons to do this. Maybe the approach will reveal new facets and spark a fresh viewpoint on the world. It could also be something that you do on a lark, just for kicks.

The beauty of flipping the script is that it can have profound outcomes and tremendous possibilities. It all depends on what you are trying to accomplish. Plus, knowing how to best carry out a flip-the-script endeavor is a vital consideration too. You can easily mess up and get nothing in return.

All of this comes to the fore for those using generative AI and wanting to make sure that they are leveraging the latest and boldest of prompt engineering capabilities and techniques.

In today’s column, I am continuing my ongoing series about advances in prompt engineering and will be taking you through a technique known as flipped interaction. Here’s the deal. You flip the script, as it were, getting generative AI to ask you questions rather than having you ask generative AI your questions.

That might seem at first glance rather surprising, and perhaps even has a hint of being nonsensical. We are all accustomed to the idea that you enter prompts to generative AI that contains burning questions you have about the world. The generative AI does its computational and mathematical machinations and produces a generated answer or response. This is the standard way of doing things.

Think of it this way:

Boom, rinse and repeat.

The flipped interaction urges you to try doing this:

Is that nutty?

Or is it perhaps ingenious?

I’ll explain that it can be ingenious for a variety of stellar reasons.

You are not alone if you’ve not tried this. Few people who use generative AI seem to realize that this prompting strategy has true value. I would dare say that most people probably have never even heard of the approach. For those people who have perchance seen or read something about the flipped interactions method, they likely got a mere morsel of what this is all about. The typical depiction gives you a paragraph or two. You are otherwise left to your own devices to figure out what flipped interaction can do and how it earns a prized spot in your plethora of prompt engineering-worthy skills.

I want to correct the slew of misimpressions about the flipped interaction method. I also want to showcase the immense potential that flipped interaction has for your daily use of generative AI. We will closely look at the tradeoffs and consider what contexts warrant a flip-the-script prompting strategy. This will require getting into the weeds and making sure you are fully aware of the ins and outs of employing flipped interaction in generative AI.

Before I dive into this in-depth exploration, let’s make sure we are all on the same page when it comes to the keystones of prompt engineering and generative AI. Doing so will put us all on an even keel.

Prompt Engineering Is A Cornerstone For Generative AI

As a quick backgrounder, prompt engineering or also referred to as prompt design is a rapidly evolving realm and is vital to effectively and efficiently using generative AI or the use of large language models (LLMs). Anyone using generative AI such as the widely and wildly popular ChatGPT by AI maker OpenAI, or akin AI such as GPT-4 (OpenAI), Bard (Google), Claude 2 (Anthropic), etc. ought to be paying close attention to the latest innovations for crafting viable and pragmatic prompts.

For those of you interested in prompt engineering or prompt design, I’ve been doing an ongoing series of insightful looks at the latest in this expanding and evolving realm, including this coverage:

Anyone stridently interested in prompt engineering and improving their results when using generative AI ought to be familiar with those notable techniques.

Moving on, here’s a bold statement that pretty much has become a veritable golden rule these days:

If you provide a prompt that is poorly composed, the odds are that the generative AI will wander all over the map and you won’t get anything demonstrative related to your inquiry. Being demonstrably specific can be advantageous, but even that can confound or otherwise fail to get you the results you are seeking. A wide variety of cheat sheets and training courses for suitable ways to compose and utilize prompts has been rapidly entering the marketplace to try and help people leverage generative AI soundly. In addition, add-ons to generative AI have been devised to aid you when trying to come up with prudent prompts, see my coverage at the link here.

AI Ethics and AI Law also stridently enter into the prompt engineering domain. For example, whatever prompt you opt to compose can directly or inadvertently elicit or foster the potential of generative AI to produce essays and interactions that imbue untoward biases, errors, falsehoods, glitches, and even so-called AI hallucinations (I do not favor the catchphrase of AI hallucinations, though it has admittedly tremendous stickiness in the media; here’s my take on AI hallucinations at the link here).

There is also a marked chance that we will ultimately see lawmakers come to the fore on these matters, possibly devising and putting in place new laws or regulations to try and scope and curtail misuses of generative AI. Regarding prompt engineering, there are likely going to be heated debates over putting boundaries around the kinds of prompts you can use. This might include requiring AI makers to filter and prevent certain presumed inappropriate or unsuitable prompts, a cringe-worthy issue for some that borders on free speech considerations. For my ongoing coverage of these types of AI Ethics and AI Law issues, see the link here and the link here, just to name a few.

With the above as an overarching perspective, we are ready to jump into today’s discussion.

The Purpose Of Flipped Interaction For Generative AI

The simplest way to think about flipped interactions is that you merely tell the generative AI to start asking you questions. Voila, you have flipped the script.

Why do something that seems entirely counterintuitive?

It does strike one as a puzzling head-scratcher.

We conventionally come to use generative AI as a kind of oracle that can spew forth impressive answers. This is certainly not because the AI app is sentient. It is because the generative AI has been vastly data-trained on human written text and garnered a huge pattern-based mimicry of what humans say. The expectation is that all of that mathematical and computational pattern matching will undoubtedly tap into human writings in a manner that can reveal interesting and insightful generated responses.

The logic for why you would use a flipped interaction is actually quite straightforward if you are willing to give the curiosity a moment of keen reflection.

Here are my six major reasons that I expound upon when conducting workshops on the best in prompt engineering:

Each of those reasons for invoking a flipped interaction deserves a brief explanation. I will do so, one at a time.

(1) Inform or data-train the generative AI

The foremost reason to flip the script or do a flipped interaction with generative AI is to inform the AI or do a bit of on-the-fly data training into the AI.

Some refer to this as “teaching” generative AI about new things that the AI app hasn’t necessarily already encountered when it was initially data trained. I have an unsettling pit in my stomach when people say that they are “teaching” generative AI, due to the potential anthropomorphizing of the AI.

You see, we usually teach sentient beings such as fellow humans and thus associate teaching with doing so for a sentient being. The problem with claiming you are teaching generative AI is that this implicitly comes across as though the AI is presumably sentient. Anyway, despite that reservation, by and large, the AI community and those outside the AI community tend to proclaim that using a flipped interaction is a form of teaching an AI app. I go along with this though my teeth are gritted.

Back to the matter at hand.

Suppose you know a lot about the dodo bird. You are a world expert on the dodo. That is about the only bird you know. While using generative AI, you discover that upon asking the AI app about the dodo, you get a generated response from the AI that it doesn’t have much if anything to indicate about the dodo. Turns out that the initial data training did not contain substantive materials about the dodo. Ergo, the generative AI has sparse content that was pattern-matched on the topic of the dodo.

You are chagrined at this. First, you believe it is a travesty that generative AI doesn’t have an in-depth capability about the dodo. Everyone and everything need to know about the dodo. Secondly, you wanted to have generative AI compare and contrast the dodo to other birds of a more common nature. This comparison cannot be undertaken because the generative AI is pretty much empty when it comes to the wonderful and greatly prized dodo.

What can you do about this dilemma and worrisome situation?

Easy-peasy is that you can inform or data train the generative AI about the dodo. After doing so, you can then switch into the usual mode of asking the generative AI about things such as comparing the dodo to other more common birds.

You decide to momentarily flip the script.

One path would be to have you type in all the fascinating and teeny tiny details about the dodo that you might have in your mind. You might do this as one gigantic and lengthy prompt. A brain dump for the generative AI.

On the other hand, this might not be especially productive. Maybe a better approach would be to get the generative AI to ask you a series of questions about the dodo. This is helpful to you since you can then merely answer the generated questions. This might also be better for the generative AI in that rather than trying to flood the AI with a bunch of random stuff about the dodo, the AI will be directing step-by-step as to what the AI seemingly needs to be data trained on about dodo birds.

You invoke the flipped interaction.

Doing so is extremely easy. You could enter a prompt that tells the AI app to start asking you questions about dodo birds. The next thing that will happen is that the AI app will pepper you with questions.

I ought to forewarn you that there is a bit more involved in that you should also clue in the AI about a variety of other particulars associated with the Q&A that you want it to undertake. We’ll get to that in the next section herein.

Meanwhile, let’s consider the additional reasons for using a flipped interaction.

(2) Discover what kinds of questions arise in a given context

Discovering the kinds of questions that arise in a given context is another great reason to use the flipped interaction technique of prompting.

Imagine that you are curious about what types of questions might be pertinent to a particular topic or subject area. For example, what kinds of questions usually arise when trying to fix the plumbing in your house? You might have at times gone to the Internet to see what questions people ask when they seek to fix their plumbing. Those questions are a handy-dandy way for you to discover what you likely might need to know or what might suddenly arise while fixing your plumbing.

You can merely tell the generative AI to hit you with the most commonly asked questions about fixing the plumbing in a house. This can be a list that the generative AI generates.

Another akin approach would be to have the generative AI ask you those questions, one at a time. This might prod you to think more closely about the matter at hand.

(3) Learn from the very act of being questioned by the AI

You can potentially learn something by the act of being questioned.

Let’s continue my above description about fixing your plumbing. You get asked a series of questions by the generative AI. For example, the AI asks you if you have the right plumbing tools. Oops, you realize that you assumed that your everyday toolbox would have all the tools you need. The generative AI has sparked your realization that there are specialized tools that plumbers use. You make a note to go to the hardware store and get the needed tools.

Getting asked a question has allowed you to learn something that otherwise you didn’t know or hadn’t considered.

(4) Allow yourself intentionally to be tested and possibly scored

I’ve predicted that people are going to eventually realize that they can use generative AI as a kind of test-taking mechanism, see my discussion at the link here and the link here.

With the flipped interaction, you tell the generative AI to test you on a particular subject or topic of interest. The AI app will ask you a series of questions. If you don’t want to be scored or assessed, you can just say that the AI should not do any rating of your answers (the usual default is that it won’t, thus, you are more likely to need to explicitly say you want to be scored, if that’s what you want).

Suppose you are in college and have a test coming up about the life of Abraham Lincoln. You have studied a bunch of written materials about his life. Are you ready for the test that you’ll be taking tomorrow? All you need to do is flip the script and tell generative AI to ask you questions about Abraham Lincoln. You can do this as much as you like, over and over, until you feel that you are adequately prepared for the test you’ll be taking.

A few quick caveats that I’ll cover more later on herein. It could be that the questions about Abraham Lincoln do not end up covering whatever the teacher comes up with. Maybe the generative AI wasn’t data-trained on certain aspects of Lincoln’s life. You cannot assume that the generative AI is all-encompassing. It isn’t. Furthermore, and quite bothersome, the generative AI might contain errors or falsehoods about Lincoln, and possibly make up stuff or do a so-called AI hallucination about Lincoln. The issue here is that when you see such a question, you might not realize that the question is erroneous. Sadly, you believe the question contains truths and you memorize facts about Lincoln that are incorrect.

Be careful and mindful when using generative AI, including when doing a flipped interaction.

(5) Do this as a game or maybe just for plain fun

Some people enjoy using generative AI for playful purposes. I’ve discussed in my columns that you might tell the AI app to pretend to be a famous character in a play or novel (have it mimic a persona), and then you joyfully interact with that character (see my analysis of this trend, at the link here).

You can use the flipped interaction for game-related purposes or just for plain fun.

Imagine that you want to play a game with the AI app and you are going to make up the rules of the game. This is not checkers, chess, or any conventionally known game. It is a game of your own devising.

How can you get the AI app up-to-speed about your made-up game?

One approach would be to tell the AI that you are going to define a game and that you have a bunch of rules that the AI needs to be data-trained on. You then instruct the AI app to ask you questions about the game. This is similar to my earlier point that you can do data training on generative AI while on the fly.

Similarly, rather than making up a game, you could tell the generative AI to start asking you questions about Abraham Lincoln. You are doing so merely for the fun of it. You aren’t in school, and you aren’t trying to prepare for a test. Instead, this is something being done for simple entertainment.

(6) Other bona fide reasons

A variety of other bona fide reasons to use the flipped interaction prompting technique are floating around. I’ve given you the top ones that are most often identified.

I’m sure you can think of more.

We need to move on and cover the crucial foundations of this promising and valuable technique.

The Foundations Of Employing The Flipped Interaction Technique

I trust that you are convinced that there are some legitimate reasons to consider using the flipped interaction prompting technique. Great, since that means we can dive deeper into the foundations underlying the technique. I most assuredly hope that you are tempted to use the flip-the-script and are now waiting with bated breath to know how to do so.

Fasten your seatbelts and get yourself ready.

Let’s envision that you tell generative AI to start asking you questions.

If that’s all that you stated, there is a tremendous amount of ambiguity at that juncture. What is the subject matter or topic underlying the questions that you want to be asked? How many questions are to be asked? Do you have a preference as to the style or phrasing of the questions? Should the answers you are giving be scored or rated?

On and on the litany of open issues goes.

I would assume that at least the idea of stating the subject matter would normally be at the top of your mind for anyone seeking to do a flipped interaction. Indeed, that is probably the shortest and most minimal of ways to use the flipped interaction. You could merely say that you want to be asked questions about dodo birds. Period, end of story.

The generative AI would likely proceed.

That being said, since generative AI is like a box of chocolates, you never know what you might get. The questions probably will be about dodo birds, though maybe not in the direction you had in mind. Perhaps the AI will ask you questions about whether dodo birds can fly a plane or pilot a rocket ship. I doubt those are the questions you assumed you would be asked.

Okay, we’ve now identified that you can invoke a flipped interaction by the simple act of telling generative AI to start asking you questions. But, if you want this to be productive, you ought to sensibly specify more about what you want to happen.

You can provide a speck of guidance to the AI or you can provide a bushel of guidance to the AI. It’s up to you. Some people aren’t excited about having to specify a bunch of stuff. They just want to get on with the show. Others are more studious and believe that by being detailed in their guidance the effort will turn out more beneficial.

I might also add that you can either specify upfront when you initiate a flipped interaction the details of what you want to do, or you can sprinkle the nuances as you go along. The beneficial thing about generative AI is that you can interactively do things. You don’t necessarily need to disgorge an entire missive at the get-go. You can do as you proceed. For myself, I admit that I lean toward trying to specify as much as possible at the get-go when invoking a flipped interaction. This seems sensible to me. I then will adjust or fine-tune as the Q&A proceeds.

Please use the flipped interaction techniques in whatever style befits your personal preferences.

Here are my twelve foundational recommended indicators that you should consider and possibly convey to the generative AI when invoking the flipped interaction prompting strategy:

I’ll quickly go over those with you.

Rather than elaborating on each one, and due to space limitations herein, I’ll cover them collectively and give just distinct highlights.

First, you need to tell the AI app that you want to flip the interaction. This is easily done. All you need to do is indicate in a prompt that you want the generative AI to start asking you questions. Voila, you are ready to roll. There is though more to the invoking of a flipped interaction. I had mentioned earlier that just saying you want to be questioned is not really sufficient since the AI app won’t be informed as to what you want to be asked about.

Your best bet is to also indicate the rules of the road about the Q&A. One key aspect will be how many questions the AI app ought to ask you. Perhaps you already know how many questions will be required and thus you can give an exact count such as five questions or twenty questions. Much of the time you probably won’t in advance know how many questions are going to be needed. You can ergo make the Q&A more open-ended and let the AI app keep asking you questions until you tell it to stop doing so.

Another aspect involves indicating when the Q&A should get underway. The default assumption usually is that you say that you want the questioning to get started and it happens right away. Sometimes you might want to momentarily delay the Q&A and do some other preparatory actions with the generative AI before getting into the questioning mode.

There is no doubt that the most significant element to specify is the topic or subject matter underlying the Q&A endeavor. You should set the context for the questioning. This can range from being extremely specific such as the eating habits of the dodo bird or might be more wide-ranging such as the overall meaning of life.

The tone of the questions can vary quite a bit. Generative AI is usually already pre-tuned to ask questions in a civil manner. If you want the questions to be harder-hitting or have some particular style, make sure to mention this when you indicate that a Q&A is to take place. The AI app will also typically default to asking you one question at a time. This is not a sure bet. If you want the Q&A on a one-at-a-time basis, say so. If you want the Q&A to consist of two questions at a time or maybe a slew of questions all at once, you can get that to happen by saying so.

When you give your answers during the Q&A, you don’t know that the AI app will necessarily be attempting to data-train on whatever you have to say. The usual default is that the generative AI will in fact be tracking your answers and be ready to at some point regurgitate or respond accordingly. I like to make this an apparent consideration and normally explicitly tell the AI app to pay attention.

A less often considered facet is whether you can challenge the questions that the AI is giving you. The typical default is that you can do so. If you get a question that you think is off-target, provide a reply saying so. I like to make sure that the AI app is ready for such challenges and therefore tend to mention as such when I initiate the flipped interaction. On a related matter, you can say whether the AI is to challenge your answers. Ordinarily, the AI won’t challenge your answers and will accept them as given. You can tell the generative AI to balk at answers that seem awry. One potential irritant is that the AI app can become ferociously antagonistic and you won’t likely enjoy the Q&A process (you’ll need to tell the AI app to tone things down if that happens).

There are some additional considerations to keep in mind. You might want to temporarily pause the Q&A. This can be done at any time by merely stating as such when responding to a question. You can also establish upfront a pause signal such as a word or catchphrase.

Some people prefer to provide their answers and have the AI app proceed immediately to the next question. For me, I like to have the generative AI echo my answers, doing so after each of the posed questions. I do this because there is always that chance that the AI app makes an error, falsehood, AI hallucination, or other mistake when interpreting your answer. I prefer to catch this straightaway and thus usually tell the AI app to echo back my answers after each question.

Suppose you answer a question and then afterward realize that you made a mistake in your answer. You can usually undo an answer by simply saying this when you are answering the next question. On an upfront basis, I sometimes indicate that I might be doing some undo actions, especially if I am unsure of the topic at hand and might be spitballing for my answers.

Finally, you can either end the Q&A by stating as such when you get the next question, or you can beforehand establish a keyword or phrase that will signal you want to stop the Q&A. At that juncture, upon ending the Q&A, it is handy to mention or state again what you want the AI app to do with the answers that you have provided. You can presumably at that point just go back into the conventional mode of asking questions and the AI answering them. I frequently explicitly state that this is going to occur, just to make sure that the AI app doesn’t go off on a tangent and starts suddenly into Q&A mode again.

Flipping out of a flipped interaction is typically done without having to outrightly say so, other than telling the AI to stop the Q&A. For an added measure of reassurance, a few words telling the AI that the Q&A has ended is probably a worthy remark to make.

Example Of Doing A Flipped Interaction

We’ve covered the basics of undertaking a flipped interaction. I would bet that you are eager to see an example.

Wonderful, since I have one ready to show you.

Let’s imagine that I am an expert in underwater basket weaving. Have you ever heard of underwater basket weaving? There is a bit of a humorous undercurrent. One viewpoint is that underwater basket weaving refers to dipping your hands into a pool of water and weaving a basket in that manner. Another more comical variant is that you are fully immersed in water and are doing basket weaving while completely underwater.

The catchphrase referring to underwater basket weaving gradually became a means of suggesting that someone is doing something of a wasteful nature or otherwise doing a task we might consider silly or unnecessary. For example, college students who take a class of a seemingly vacuous nature are said to be learning how to do underwater basket weaving.

For purposes of carrying out a Q&A with generative AI on a flipped interaction basis, I logged into ChatGPT and tried to seriously engage in a discussion about underwater basket weaving. Admittedly, this is a bit tongue-in-cheek. A human might have right away questioned my sincerity. By and large, generative AI will usually let you lead on such aspects and accept that you are serious and that whatever topic you have chosen is equally of a serious nature.

Here we go.

My prompting setup for getting the flipped interaction underway is this:

Notice that I weaved into my prompt the various parameters of the flipped interaction. I didn’t exhaustively cover every possible consideration. I gave just enough that I thought would ensure that the Q&A would proceed sensibly. I gave the topic to be covered, I indicated I wanted to be asked five questions, doing so on a one-at-a-time basis. I also instructed the generative AI to do a recap at the end. I also mentioned that I might end the Q&A sooner than five questions.

ChatGPT generated this response:

We are now off to the races and ChatGPT has asked me the first of the presumed five questions. I had no idea what questions I would be asked, other than that I had specified they were about underwater basket weaving.

My prompt as a reply to the posed AI question was this:

ChatGPT then generated this response to my first answer:

I next provide a prompt that answers the question that ChatGPT posed in the above-generated response:

ChatGPT generates a response indicating I have provided an answer to that question and then poses the next question for me to answer:

I provided a prompt that gave my answer to the question, and also decided that I had enough of the Q&A and wanted to stop the flipped interaction:

The generated response by ChatGPT said this:

Observe that ChatGPT provided a recap of my answers. This was due to the aspect that I had asked for the AI app to do so, having instructed that this is supposed to occur (mentioned in my initial setup prompt for the Q&A). The chances are that the generative AI might have given me a recap anyway, but I wanted to ensure that I did indeed get one.

The recap looks reasonably on par with my answers. Thankfully so. You cannot assume that the generative AI would have accurately or aptly absorbed the answers. I used the recap to double-check and try to ascertain whether any AI errors, glitches, AI hallucinations, or other misinterpretations occurred. For now, it looks good.

I next thought that it might be advantageous to do an additional double-check by explicitly asking ChatGPT about underwater basket weaving. This will give me additional reassurance that the generative AI likely suitably got data-trained on my answers. With the flipped interaction over, I asked this question to ChatGPT:

ChatGPT generated this response:

The generated response goes quite a bit further than my earlier answer about the equipment involved in underwater basket weaving. This shows you that your answer to the posed questions will undoubtedly get embellished and enmeshed into other data that the generative AI has already pattern-matched on. Depending on how things go, this can be a good thing or a bad thing. In this case, the elaboration about the scuba equipment and the snorkeling equipment seems okay.

I asked several additional questions, and the answers were of an akin caliber, namely that ChatGPT was able to use my answer and tended to elaborate on what I had said.

One special aspect to consider is the temporary nature of the data training that you might have done when doing a Q&A during a particular conversation with the generative AI. Keep in mind that this might only last for the course of the specific conversation underway with the generative AI.

Here’s the rub. An AI maker can choose whether to have their generative AI data train on individual conversations or just allow a given conversation to come and go. Sometimes an AI maker might want to permanently incorporate the new data training into the AI, while other times they might not want it to happen (such as a dialogue that is untoward or has other adverse properties).

Conclusion

Congratulations, you are now ready to astutely use flipped interactions.

I urge you to try out this handy capability of generative AI. Your prompt engineering prowess should include the ability to skillfully wield flipped interactions, including knowing how to best invoke a flip of the script and how to squeeze as much value out of doing so.

For those of you who want to get even more advanced with flipped interactions, here are some of the exercises that I do in my workshops on prompt engineering:

Each of those variations has a useful purpose. Explore the flipped interaction possibilities and be ready to invoke a flip-the-script whenever suitable.

A final remark for now on this weighty inverse-of-control matter.

We are often told to avoid falling into a rut in life. The same can be said about those who frequently use generative AI. The everyday common rut is that you ask questions of the AI and wait to see the generative answers.

Time to flip the script!

Think about how you can employ flipped interactions when using generative AI. Get yourself out of an altogether mindless rut and leverage these undervalued and underused prompt engineering techniques. You’ll be a better person (well, maybe so) and assuredly a better consumer of generative AI for having done so.

BattleRoyal Hack: Threat Actors Use 'Special' Type of Social Engineering to Deliver DarkGate RAT Malware

In a sweeping wave of cyber onslaughts this fall, a threat actor, dubbed "BattleRoyal," executed a myriad of sophisticated social engineering campaigns against organizations in the United States and Canada. 

The primary objective behind these orchestrated attacks was to infiltrate systems with the elusive and multifaceted DarkGate malware. 

Researchers from Proofpoint uncovered a web of diverse tactics and strategies deployed by this elusive entity.

(Photo : Gerd Altmann from Pixabay)With hackers innovating their social engineering techniques to launch malware attacks, the same thing goes with "BattleRoyal" cyber criminals who are using some new methods to infiltrate systems.

The researchers, in a recent blog post, grapple with the challenge of definitively categorizing BattleRoyal as either a completely new threat actor or an entity linked to existing ones. 

The complexity arises from the extensive range of tactics, techniques, and procedures (TTPs) employed by BattleRoyal, making it a formidable and elusive adversary in the cybersecurity landscape.

According to DarkReading, BattleRoyal employs an array of techniques to deploy DarkGate, coupled with the recent inclusion of the NetSupport remote control software. The arsenal includes phishing emails on a large scale, fake browser updates, exploitation of traffic distribution systems (TDSs), malicious VBScript, steganography, and exploitation of a Windows Defender vulnerability. 

Intriguingly, despite the diversity of these tactics, there is no documented instance of successful exploitations thus far.

Related Article: Lazarus Group Still Exploits Log4Shell: What Are Andariel's Recent Cyberattacks?

BattleRoyal demonstrates a penchant for traditional email phishing, orchestrating over 20 campaigns between September and November. These campaigns, comprising tens of thousands of emails, often commence with seemingly innocuous messages. 

The inclusion of multiple traffic distribution systems (TDSs) in the embedded links serves as a common tool for redirecting users to URLs exploiting CVE-2023-36025.

Speaking of the exploit, the attackers leverage CVE-2023-36025, a critical bypass vulnerability with a score of 8.8, to compromise Microsoft Defender SmartScreen. 

Irony ensues as SmartScreen, designed to enhance security by preventing users from falling prey to phishing sites, becomes a gateway for BattleRoyal's malicious activities. It is noteworthy that BattleRoyal potentially exploited this vulnerability as a zero-day, preceding its public disclosure.

At the heart of BattleRoyal's intricate web lies DarkGate, a malware-encompassing loader, cryptominer, and remote access Trojan (RAT) capabilities. Although DarkGate has existed for over half a decade, its resurgence in October marked a significant uptick in activity. The surge is attributed to the malware developer leasing it to a select group of affiliates, a practice advertised on cybercriminal hacking forums. 

The malware's recent switch to NetSupport prompts speculation about BattleRoyal's motivations and the ever-evolving landscape of cybersecurity threats.

"Proofpoint regularly sees TDSs used by threat actors in attack chains, specifically cybercrime campaigns. Threat actors use them to ensure the computers they want to be compromised are, and anything that doesn't meet their standards such as a bot, possible researcher, etc., will be redirected away from payload delivery. The two most common TDSs these days are the same ones used by BattleRoyal: 404 TDS, and the legitimate Keitaro TDS," Proofpoint's senior threat intelligence analyst Selena Larson said.

RAT malware is everywhere and they come in various forms just like other malicious software. 

For instance, a new breed of the threat called "SugarGh0st" RAT was deployed by alleged state-sponsored hackers from China. The campaign was created to attack Uzbekistan and South Korea.

Read Also: Researchers Discover Malicious Chrome Extensions Disguised as Fake VPN

ⓒ 2023 TECHTIMES.com All rights reserved. Do not reproduce without permission.

Proofpoint Exposes Sophisticated Social Engineering Attack on Recruiters That Infects Their Computers With Malware

A new report from U.S.-based cybersecurity company Proofpoint exposes a new attack campaign operated by a financially-oriented threat actor dubbed TA4557 with high financial data theft risks and possibly more risks such as intellectual property theft.

In this social engineering campaign, the threat actor targets recruiters with benign content before infecting their machines with the More_Eggs malware. This threat actor takes extra care to avoid being detected.

Jump to:

The latest attack campaign from threat actor TA4557, as exposed by Proofpoint, targets recruiters by sending them a direct email. The group pretends to be an individual interested in a job (Figure A).

Figure A

The email does not include any malicious content. Once the recruiter replies to the email, the attacker replies with a link leading to an attacker-controlled website posing as an individual’s resume (Figure B).

Figure B

An alternative method used by the threat actor consists of replying to the recruiter with a PDF or Microsoft Office Word file containing instructions to visit the fake resume website.

The website employs filtering mechanisms to assess whether the subsequent phase of the attack should be initiated. If the criteria for filtering are not met, the user is presented with a plain text resume. If the filtering checks are successfully passed, the user is redirected to the candidate website, where they are prompted to solve a CAPTCHA.

Upon successful completion, the user is provided with a ZIP file that includes a Microsoft Windows shortcut (LNK) file. If executed, the LNK file abuses a legitimate software ie4uinit.exe to download and run a scriptlet from a location stored in the ie4uinit.inf file. The technique known as Living Off The Land consists of using existing legitimate software and tools to accomplish malicious actions on the system, which minimizes the likelihood of being detected.

The downloaded scriptlet decrypts and drops a DLL file before attempting to create a new process to execute the DLL by using Windows Management Instrumentation. If it fails, the scriptlet tries another approach by using the ActiveX Object Run method.

Once the DLL is executed, it decrypts the More_Eggs malware along with the legitimate MSXSL executable. Then, it initiates the creation of the MSXSL process using the WMI service. The DLL deletes itself once the infecting process is done.

According to Proofpoint, More_Eggs is a malware that enables persistence and profiling of the infected system; it is also often used to download additional payloads.

TA4557 employs various strategies to evade detection and maintain a low profile, demonstrating a commitment to staying below the radar.

In other attack campaigns, mostly in 2022 and 2023, the threat actor used a different technique that mainly consisted of applying for open positions on job offer websites. The threat actor used malicious URLs or files containing malicious URLs in the application, but the URLs were not hyperlinked, meaning the recipient had to copy and paste the URLs directly into their browser. That technique is interesting because the use of such a link will likely not trigger as many security alarms. According to Proofpoint researchers, TA4557 still uses that technique in parallel with the newly reported technique.

In addition, the threat actor previously created fake LinkedIn profiles, pretending to be a recruiter and reaching out to people looking for a job.

The use of LOTL techniques is an indication that the threat actor tries to stay discreet and undetected.

The DLL file used by the threat actor employs anti-sandbox and anti-analysis techniques, such as incorporating a loop strategically crafted to extend the execution time while slowly retrieving the RC4 key needed to decipher the More_Eggs backdoor. Multiple checks are also done to see if the code is running in a sandbox or in a debugging environment. Once the infection process has gone through, the DLL deletes itself to remove evidence of its presence and render incident analysis harder.

TA4557 is described by Proofpoint as a “skilled, financially motivated threat actor” who demonstrates sophistical social engineering. The group regularly changes its sender emails, fake resume domains and infrastructure. Proofpoint believes the same threat actor targeted anti-money laundering officers at U.S. credit unions in 2019.

From a global point of view, the researchers noticed an increase in threat actors engaging their targets using benign content first to build confidence during the interaction before sharing harmful content.

TA4557 uses social engineering to infect the machines of unsuspecting victims, which are recruiters in this attack campaign; in the past, the threat actor also targeted individuals looking for jobs. So, it is advised to educate all people involved in hiring processes about these kinds of social engineering techniques.

It is recommended never to open a document or click on a link that seems suspicious. When in doubt, employees must alert their IT department and have the documents or links analyzed.

Security solutions must be deployed on all endpoints, and alerts should be carefully analyzed.

Email content should be analyzed by security solutions capable of detecting anomalies instead of only URLs or attached files to try to detect social engineering-based campaigns.

All operating systems and software must be kept up to date and patched to avoid being compromised by common vulnerabilities.

Disclosure: I work for Trend Micro, but the views expressed in this article are mine.