Magazine menu Close Magazine menu

Search results

Showing 54 results

adverserial AI

How to keep a business safe from attacks through AI’s machine-learning

  • Artificial intelligence
  • min read
    • We hear every day of new uses companies find for artificial intelligence (AI,) but security of machine-learning models that underpin AI doesn't get as much attention. When machine learning goes wrong, it's never good for business – like when a text-generation model spews racists slurs or social media comment filters are tricked into displaying toxic comments. But some businesses are taking action – especially those at the top of their game. The cybersecurity community is producing guidelines like MITRE ATLAS to help businesses to secure their machine learning models and tech giants like Microsoft and Meta are assembling expert teams to safeguard the machine learning beneath their business-critical AI. One way Kaspersky protects the machine-learning technologies powering everything from our malware detection to antispam, is we attack our own models. Machine learning is vulnerable because it learns like us Known threats to machine-learning models, called adversarial examples, don't stem from error or misconfiguration – they're part of how the machine-learning model is made. Alexey Antonov, Lead Data Scientist at Kaspersky and expert in machine learning-based malware detection, has a great way to describe these threats. Antonov's team set out to attack their own malware detection model. Rather than being programmed with rules, this model learns to tell malicious from legitimate software by training on troves of malware examples collected over the years. It's like how our minds learn – and like our minds, it can be fooled. "We craft specific data to confuse the algorithm", says Antonov. "For example, you can glue pieces of data to a malicious file until the model stops recognizing it as malicious." Unlike bugs in traditional software, adversarial examples are hard to fix. There's not yet a universal way to protect against them, but you can improve your machine learning's security by adding adversarial examples to training data. Diluting the impact of poisoned data Machine-learning models are so called because they aim to model what happens in the real world. But what they really do, is describe the data used to train them mathematically. It can produce biased results if the training data is biased, for example, a face-recognition model trained predominantly on white faces will struggle to recognize People of Color. If an adversary can modify your training data (for example, if you use openly available datasets,) they will change your models – known as 'data poisoning.' Nikita Benkovich, Head of Technology Research at Kaspersky, says his team realized a model protecting enterprises from spam could be attacked with data poisoning. Models are frequently retrained because spam is always evolving, so an adversary could send spam emails using a legitimate company's technical email headers, perhaps causing the model to stop all customers receiving that company's real emails. "We had many questions," says Benkovich. "Can you actually do it? How many emails would we need? And can we fix it?" After verifying such an attack was possible, they looked at ways to protect the system, coming up with a statistical test that would flag anything suspect. How businesses can prevent adversarial attacks on AI Adversarial attacks can affect fields as broad as object detection to machine translation, but Alexey says real-world adversarial attacks aren't yet happening on a large scale. "These attacks need highly-skilled data scientists and much effort, but they can be part of Advanced Persistent Threats (APTs) for targeted attacks. Also, if a security solution relies solely on machine learning, an adversarial attack can be highly profitable – fooling the algorithm once, the adversary can use the same method to make new strains of malware the algorithm can't detect. That's why we use a multi-layered approach." Benkovich says pay close attention to where your machine-learning training data comes from, or 'data provenance.' "Know where your training sample comes from." Both experts agree the best way to keep your models protected is to test by attacking them yourself before others do. Alexey quotes Chinese military strategist Sun Tzu for the best advice: "If you know the enemy and know yourself, you need not fear the result of a hundred battles."
Human paints AI with human features, while AI paints human with machine features

Reaching the technological singularity: What will happen when machines become smarter than us?

  • Future tech
  • min read
    • Picture a world where machines have long since passed the Turing Test. When people can't tell the difference between artificial and human intelligence. In this world, will humans become obsolete, like the floppy disk? Or is it a world where artificial and human intelligence merge to usher in the next step of our evolution? That's the technological singularity, the theoretical time in the future where technological development outpaces our abilities to maintain control over it. For better or worse, it's undoubtedly something that will result in never seen changes to our world. When will we reach technological singularity? Join any discussion about business technology, and it won't be long before someone mentions artificial intelligence (AI). The amount of digital data in the world has already surpassed our abilities to manage it, a development, which has made clear the need for algorithms to do the job for us. But algorithms and AI aren't the same things. They're not even close. An algorithm is a computer program that parses data from sets too large for human interpretation, whereas true AI is capable of thinking for itself, making its own decisions, and having free will. True artificial intelligence doesn't exist. Yet. The explosion of AI has long been a popular trope of science fiction, but the reality is edging ever closer. Ray Kurzweil, world-renowned futurist and Google's Director of Engineering, believes technological singularity is going to happen before 2045. Take note: so far, an impressive 86 percent of his predictions made since the '90s have become a reality. But even if that sounds optimistic (or pessimistic, depending on how you look at it), it's the logical next step in the evolution of technology. Once computers can learn for themselves without being taught and trained with data that we've collected, we can expect profound society changes on a scale never before seen. How can we keep up with technology? The emergence of an artificial superintelligence would bring us greater inventive and problem-solving skills than humans are capable of. This might lead to the creation of a new 'species,' one which might not necessarily have human interests at its heart. But so far, we're nowhere near developing machines with human-level intelligence and the ability to make decisions independently. Now we can enhance human intelligence and capabilities with tools and technology. Indeed, that's precisely what we've been doing for hundreds of thousands of years, ever since our ancestors figured out how to light fires. But today, the stakes are much higher with projects like brain-to-computer interfaces as active areas of research. This may, in turn, lead to radical developments like mind uploading, in which you entire consciousness can be uploaded to and stored on a machine. At least, it could be useful for those millennia-long interstellar voyages. Such developments might sound dystopian since they have the potential to transform what it means to be human ultimately. At the same time, the convergence of technology and humanity, an area of study known as transhumanism, appears to be the only way we can keep up with the constant pace of change. It's either that or the machines eventually outpace our natural evolution and take over. The survival instinct If we can't keep up, and technological evolution ends up outpacing our capability of keeping control over it, then we could face profound risks to security. Each of the seven characteristics that define every biological organism comes together to form a singular goal – survival. Technological singularity occurs when AI matches our desire and capability to survive as a species, and there's little reason to think that AI would have any evolutionary motivation to be friendly to humans. Today, machines are programmed by us to create the outcomes we want. But what happens when they're capable of programming themselves? Time for Humanity 2.0? Imagine you're an expert on machine learning, working on an artificial intelligence algorithm that will be able to create other AIs by itself. Are you merely training your replacement, or have you become an architect of your obsolescence? Welcome to Humanity 2.0, where the human condition is no longer about our biological form. It's a time when machines become an integral part not just of our societies, but also ourselves. Conversely, this could also lead to newer and deeper social class divisions than we've ever seen, as people are separated into those who are augmented by technology and those who are not. There's a safety benefit of not being physically augmented by technology – no one will be able to hack into your mind. How humans could remain our biggest threat Perhaps it isn't machines that we should be worrying about at all. Assuming we'll always be able to control artificial intelligence, we could end up transforming it into a deadly weapon. It's already happening. The proliferation of social media, for example, is one of the most significant security and privacy concerns this century. That's not so much down to vulnerabilities in the technology, but because it's a highly effective medium for conducting social engineering attacks. Coupled with the rise of AI, cybercriminals will be better equipped to analyze social media conversations to emulate users' writing styles and craft more convincing messages to their victims. Even before we reach the era of true AI, the ability of algorithms to parse enormous amounts of data is already misused. On the one hand, algorithms can perform many useful tasks to improve our world, such as climate modeling and disaster prediction, while things like neural interfaces can help victims of spinal cord injuries. On the other, the ability of humans to interfere with information systems makes them a more significant threat than the technology itself. Perhaps it would be better if we just let the machines take over in the hope they'd nurture us like the irresponsible children we are. To fear or not to fear Ray Kurzweil has a more positive outlook than most about the singularity. He claims it's an opportunity for humankind to improve by making us smarter and better at all the things we value. Other science and technology leaders, including Elon Musk, Bill Gates and the late Stephen Hawking, are not so sure. But one thing's for sure – there will come a time when our augmented descendants look back on us as rather quaint and exotic creatures. While we're still decades away from the singularity, it presents itself as the next big step in the evolution of technology. Emerging tech, like autonomous vehicles and facial recognition, are already paving the way. True artificial intelligence is coming sooner or later, which is why it's time for the business world to start taking AI ethics and security issues seriously. It's no longer the concern of generations far in the future. This article reflects the opinion of its author.
Fast Forward Can machines outsmart us

Can machines outsmart us?

  • min read
    • Are we capable of telling machines apart from humans? Some evidence suggests we're surprisingly easy to fool. And now with the advanced ways machines are learning and their startling results, we must ask questions about our relationship with artificial intelligence of the future. Listen to Fast Forward audio documentary episode 4. Explore the Fast Forward series with episode guides and interview with experts.

Can machines outsmart us?

  • min read
    • Are we capable of telling machines apart from humans? Some evidence suggests we're surprisingly easy to fool. And now with the advanced ways machines are learning and their startling results, we must ask questions about our relationship with artificial intelligence of the future. Listen to Fast Forward audio documentary episode 4. Explore the Fast Forward series with episode guides and interview with experts.
insight story ai machine learning

How do you make AI work for you?

  • min read
    • Effective use of artificial intelligence (AI) and machine learning can mean a huge competitive advantage for your business. But a recent survey found that only 10 percent of companies using AI were getting decent returns on their investment. How do you make sure you get the most out of this powerful and fast-developing tech? We have some answers.
Robot fires lasers at a computer network, obliterating it while another robot watches

Predictions for Earth 2050: How will cybersecurity change?

  • Cybersecurity
  • min read
    • If you're like me, having a long-term vision is a wonderful aspiration… frequently quashed by today's more screamingly imminent 'to do' list. Five-year plan?  I'm still working on the next five days. But the next five minutes? That's a no-brainer: let's get immersed in the futuristic predictions for the world in 2050, and the evolving role of cybersecurity in it. Cyber-immunity will intuitively protect our systems Predicted by: Eugene Kaspersky, CEO, Kaspersky In 2050, digital systems have moved up to a whole new level. They're no longer assistants making life easier for us mere mortals, they've become the basis of civilization – the very framework keeping the world functioning properly. This quantum leap forward has generated new requirements for the reliability and stability of artificial intelligence (AI). Since the romantic era around the turn of the 21st century, cyberattacks are not yet extinct, but they're now only a threat to outliers who reject modern standards of digital immunity. One of the pillars of cyber-immunity is digital intuition, the ability of AI systems to make the right decisions in conditions where the source data is insufficient to make a rational choice. There's no mysticism here: digital intuition is the logical continuation of machine learning. When the number and complexity of related self-learning systems exceeds a certain threshold, the quality of decision-making rises to a level that eludes rational understanding. An intuitive solution results from the superimposition of the experience of a huge number of machine-learning models, much like the result of the calculations of a quantum computer. Digital intuition – with its ability to instantly, and correctly, respond to unknown challenges –has helped build the digital security standards of this new technological era. The robot servants are uprising Predicted by: Vitaly Kamluk, Director, Global Research & Analysis Team, Kaspersky Busy workers rejoice! Help for your home is arriving. A robot servant will cook your meals, do your laundry and clean your apartment. These autonomous systems will be able to carry out many complex tasks using the processing power of a central intelligent system, and in tricky situations, they'll switch to remote technical support managed by real people. There will be mass attempts to hack these helpful robots by labor rights activists and criminals trying to blackmail their owners. Since the servants will all be equipped with cameras, microphones and sensors, they offer a unique viewing platform for home-based espionage. Expect compromises of the manufacturers' infrastructure and the implantation of malicious code in devices with the aim of enacting, en masse, attacks that could be serious. The man(less) mining machine Predicted by: Andrey Nikishin, formerly Future Technologies Project Director, Kaspersky Mining is a fully automated process; people are not needed and former miners regain their health. One robot extracts rock from the soil, the second cleans it, the third guards the perimeter, while the fourth delivers the purified ore to industrial consumers. A hacker who gains access to the system can re-program individual robots so they take a small amount of the platinum ore to an unauthorized place. The guard robots are also hacked, so they're unaware of this theft. The big reveal of dynamic clothes Predicted by Vladislav Biryukov, Head of Consumer Social Media, Kaspersky Dynamic clothing is à la mode. The fabric's characteristics – color, pattern, lighting, transparency, and in advanced models, even the cut and length – can be automatically changed. Some people now wear shirts or dresses with variable transparency levels, which allows them to reveal more at any moment they choose. Few people 'program' their clothes by themselves; most enjoy free fashion patterns or, for the fashionistas, expensive programs from well-known designers. Initially lacking security, the process of getting a smart clothing upgrade has led to the emergence of a new kind of extortion. "Heading to a business meeting, sir? Transfer 10 bitcoins to this account number now, or we'll expose you!". The naked truth in the boardroom. Awkward. Not quite the transparency in business we were hoping to achieve in 2050. A quantum (computing) leap threatens cryptography Predicted by Alexey Malanov Cryptography is widely used in many spheres of online activity. It's no exaggeration to say that the world will experience a huge shock if suddenly everything that's encrypted today is unlocked. How cryptography works: some operations can be easily carried out in one direction, but they're very difficult to reverse. For example, an image can be quickly transformed into a picture puzzle, but to arrange all of the pieces back into the picture takes much longer. Modern public-key cryptography is based on this principle: I know two numbers, I have multiplied them and told everyone the answer. Now everyone can encrypt a message for me using the result of the multiplication, but only I can decipher this using the original multipliers. There are already quantum computers that allow you to quickly solve problems that seem to be circuit-breaking for conventional computers, like how to de-compose a composite number. The only obstacle to the end of the digital world – let's call it 'cryptocalypse' – is that quantum computers can only work with small numbers. But by 2050, quantum computing will allow security services or master criminals to access universal decryption, and they may choose to do bad things with it. For example, if a villain steals your credit card, he or she will immediately know your PIN. The ability to decipher brings with it the opportunity to forge: an ATM informs the bank one thousand dollars are withdrawn, but the villain falsifies the message to say only 10 dollars were taken out. We will need to develop new safety standards – fast. On the bright side, encryption helps with verification so you can be sure you're actually on your real bank's website, not a cleverly cloned fake, when you wire the ransom to keep your business suit from turning into your birthday suit. Phew! One challenge solved.
by Innovating leaders.
Innovative tech.
About Secure Futures

Secure Futures magazine is your go-to business guide for opinions, trends and insight into the world of technology and cybersecurity. We help your business to bring on the future. Brought to you by Kaspersky, the global cybersecurity experts.

  • For all other countries