Latest news with #SergeyLozhkin
Khaleej Times
4 days ago
- Business
- Khaleej Times
Artificial Intelligence in cybersecurity: savior or saboteur?
Artificial intelligence has rapidly emerged as both a cornerstone of innovation and a ticking time bomb in the realm of cybersecurity. Once viewed predominantly as a force for good, enabling smarter threat detection, automating incident responses, and predicting attacks before they happen — AI has now taken on a double-edged role. The very capabilities that make it invaluable to cybersecurity professionals are now being exploited by cybercriminals to launch faster, more convincing, and more damaging attacks. From phishing emails indistinguishable from real business correspondence to deepfake videos that impersonate CEOs and public figures with chilling accuracy, AI is arming attackers with tools that were previously the stuff of science fiction. And as large language models (LLMs), generative AI, and deep learning evolve, the tactics used by bad actors are becoming more scalable, precise, and difficult to detect. 'The threat landscape is fundamentally shifting,' says Sergey Lozhkin, Head of the Global Research & Analysis Team for the Middle East, Türkiye, and Africa at Kaspersky. 'From the outset, cybercriminals began using large language models to craft highly convincing phishing emails. Poor grammar and awkward phrasing — once dead giveaways are disappearing. Today's scams can perfectly mimic tone, structure, and professional language.' But the misuse doesn't stop at email. Attackers are now using AI to create fake websites, generate deceptive images, and even produce deepfake audio and video to impersonate trusted figures. In some cases, these tactics have tricked victims into transferring large sums of money or divulging sensitive data. According to Roland Daccache, Senior Manager – Sales Engineering at CrowdStrike MEA, AI is now being used across the entire attack chain. 'Generative models are fueling more convincing phishing lures, deepfake-based social engineering, and faster malware creation. For example, DPRK-nexus adversary Famous Chollima used genAI to create fake LinkedIn profiles and résumé content to infiltrate organisations as IT workers. In another case, attackers used AI-generated voice and video deepfakes to impersonate executives for high-value business email compromise (BEC) schemes.' The cybercrime community is also openly discussing how to weaponize LLMs for writing exploits, shell commands, and malware scripts on dark web forums, further lowering the barrier of entry for would-be hackers. This democratisation of hacking tools means that even novice cybercriminals can now orchestrate sophisticated attacks with minimal effort. Ronghui Gu, Co-Founder of CertiK, a leading blockchain cybersecurity firm, highlights how AI is empowering attackers to scale and personalize their strategies. 'AI-generated phishing that mirrors human tone, deepfake technology for social engineering, and adaptive tools that bypass detection are allowing even low-skill threat actors to act with precision. For advanced groups, AI brings greater automation and effectiveness.' On the technical front, Janne Hirvimies, Chief Technology Officer of QuantumGate, notes a growing use of AI in reconnaissance and brute-force tactics. 'Threat actors use AI to automate phishing, conduct rapid data scraping, and craft malware that adapts in real time. Techniques like reinforcement learning are being explored for lateral movement and exploit optimisation, making attacks faster and more adaptive.' Fortifying Cyber Defenses To outsmart AI-enabled attackers, enterprises must embed AI not just as a support mechanism, but as a central system in their cybersecurity strategy. 'AI has been a core part of our operations for over two decades,' says Lozhkin. 'Without it, security operations center (SOC) analysts can be overwhelmed by alert fatigue and miss critical threats.' Kaspersky's approach focuses on AI-powered alert triage and prioritisation through advanced machine learning, which filters noise and surfaces the most pressing threats. 'It's not just about automation — it's about augmentation,' Lozhkin explains. 'Our AI Technology Research Centre ensures we pair this power with human oversight. That combination of cutting-edge analytics and skilled professionals enables us to detect over 450,000 malicious objects every day.' But the AI evolution doesn't stop at smarter alerts. According to Daccache, the next frontier is agentic AI — a system that can autonomously detect, analyze, and respond to threats in real time. 'Traditional automation tools can only go so far,' Daccache says. 'What's needed is AI that thinks and acts — what we call agentic capabilities. This transforms AI from a passive observer into a frontline responder.' CrowdStrike's Charlotte AI, integrated within its Falcon platform, embodies this vision. It understands security telemetry in context, prioritises critical incidents, and initiates immediate countermeasures, reducing analyst workload and eliminating delays during high-stakes incidents. 'That's what gives defenders the speed and consistency needed to combat fast-moving, AI-enabled threats,' Daccache adds. Gu believes AI's strength lies in its ability to analyze massive volumes of data and identify nuanced threat patterns that traditional tools overlook. 'AI-powered threat detection doesn't replace human decision-making — it amplifies it,' Gu explains. 'With intelligent triage and dynamic anomaly detection, AI reduces response time and makes threat detection more proactive.' He also stresses the importance of training AI models on real-world, diverse datasets to ensure adaptability. 'The threat landscape is not static. Your AI defenses shouldn't be either,' Gu adds. At the core of any robust AI integration strategy lies data — lots of it. Hirvimies advocates for deploying machine learning models across SIEM (Security Information and Event Management) and SOAR (Security Orchestration, Automation, and Response) platforms. 'These systems can correlate real-time threat intelligence, behavioral anomalies, and system events to deliver faster, more precise responses,' he says. 'Especially when it comes to detecting novel or stealthy attack patterns, machine learning makes the difference between catching a threat and becoming a headline.' Balancing Innovation with Integrity While AI can supercharge threat detection, response times, and threat simulations, it also brings with it the potential for misuse, collateral damage, and the erosion of privacy. 'Ethical AI use demands transparency, clear boundaries, and responsible data handling,' says Lozhkin.'Organisations must also ensure that employees are properly trained in the safe use of AI tools to avoid misuse or unintended exposure to threats.' He highlights Kaspersky's Automated Security Awareness Platform, which now includes dedicated sections on AI-assisted threats and responsible usage, reflecting the company's commitment to proactive education. When AI is deployed in red teaming or simulated cyberattacks, the risk matrix expands. Gu warns that AI systems, if left unchecked, can make decisions devoid of human context, potentially leading to unintended and widespread consequences. 'Ethical AI governance, robust testing environments, and clearly defined boundaries are essential,' he says, underlining the delicate balance required to simulate threats without crossing into unethical territory. Daccache emphasises the importance of a privacy-first, security-first approach. 'AI must be developed and operated with Privacy-by-Design and Secure-by-Design principles,' he explains. 'This extends to protecting the AI systems themselves — including their training data, operational logic, and outputs—from adversarial manipulation.' Daccache also points to the need for securing both AI-generated queries and outputs, especially in sensitive operations like red teaming. Without such safeguards, there's a real danger of data leakage or misuse. 'Transparency, accountability, and documentation of AI's capabilities and limitations are vital, not just to build trust, but to meet regulatory and ethical standards,' he adds. Despite AI's growing autonomy, human oversight remains non-negotiable. 'While AI can accelerate simulations and threat detection, it must be guided by skilled professionals who can interpret its actions with context and responsibility,' says Daccache. This human-AI collaboration ensures that the tools remain aligned with organisational values and ethical norms. Hirvimies rounds out the conversation with additional cautionary notes: 'Privacy violations, data misuse, bias in training datasets, and the misuse of offensive tools are pressing concerns. Transparent governance and strict ethical guidelines aren't optional, they're essential.' Balancing the Equation While AI promises speed, scale, and smarter defense mechanisms, experts caution that an over-reliance on these systems, especially when deployed without proper calibration and oversight — could expose organisations to new forms of risk. 'Absolutely, over-reliance on AI can backfire if systems are not properly calibrated or monitored,' says Lozhkin. 'Adversarial attacks where threat actors feed manipulated data to mislead AI are a growing concern. Additionally, AI can generate false positives, which can overwhelm security teams and lead to alert fatigue. To avoid this, companies should use a layered defence strategy, retrain models frequently, and maintain human oversight to validate AI-driven alerts and decisions.' This warning resonates across the cybersecurity landscape. Daccache echoes the concern, emphasising the need for transparency and control. 'Over-relying on AI, especially when treated as a black box, carries real risks. Adversaries are already targeting AI systems — from poisoning training data to crafting inputs that exploit model blind spots,' he explains. 'Without the right guardrails, AI can produce false positives or inconsistent decisions that erode trust and delay response.' Daccache stresses that AI must remain a tool that complements — not replaces—human decision-making. 'AI should be an extension of human judgement. That requires transparency, control, and context at every layer of deployment. High-quality data is essential, but so is ensuring outcomes are explainable, repeatable and operationally sound,' he says. 'Organisations should adopt AI systems that accelerate outcomes and are verifiable, auditable and secure by design.' Gu adds that blind spots in AI models can lead to serious lapses. 'AI systems are not infallible,' he says. 'Over-reliance can lead to susceptibility to adversarial inputs or overwhelming volumes of false positives that strain human analysts. To mitigate this, organizations should adopt a human-in-the-loop approach, combine AI insights with contextual human judgment, and routinely stress-test models against adversarial tactics.' Gu also warns about the evolving tactics of bad actors. 'An AI provider might block certain prompts to prevent misuse, but attackers are constantly finding clever ways to circumvent these restrictions. This makes human intervention all the more important in companies' mitigation strategies.' Governing the Double-Edged Sword As AI continues to embed itself deeper into global digital infrastructure, the question of governance looms large: will we soon see regulations or international frameworks guiding how AI is used in both cyber defense and offense? Lozhkin underscores the urgency of proactive regulation. 'Yes, there should definitely be an international framework. AI technologies offer incredible efficiency and progress, but like any innovation, they carry their fair share of risks,' he says. 'At Kaspersky, we believe new technologies should be embraced, not feared. The key is to fully understand their threats and build strong, proactive security solutions that address those risks while enabling safe and responsible innovation.' For Daccache, the focus is not just on speculative regulation, but on instilling foundational principles in AI systems from the start. 'As AI becomes more embedded in cybersecurity and digital infrastructure, questions around governance, risk, and accountability are drawing increased attention,' he explains. 'Frameworks like the GDPR already mandate technology-neutral protections, meaning what matters most is how organizations manage risk not whether AI is used.' Daccache emphasises that embedding Privacy-by-Design and Secure-by-Design into AI development is paramount. 'To support this approach, CrowdStrike offers AI Red Teaming Services, helping organisations proactively test and secure their AI systems against misuse and adversarial threats. It's one example of how we're enabling customers to adopt AI with confidence and a security-first mindset.' On the other hand, Gu highlights how AI is not only transforming defensive mechanisms but is also fuelling new forms of offensive capabilities. 'As AI becomes integral to both defence and offense in cyberspace, regulatory frameworks will be necessary to establish norms, ensure transparency, and prevent misuse. We expect to see both national guidelines and international cooperation similar to existing cybercrime treaties emerge to govern AI applications, particularly in areas involving privacy, surveillance, and offensive capabilities.' Echoing this sentiment, Hirvimies concludes by saying that developments are already underway. 'Yes. Regulations like the EU AI Act and global cyber norms are evolving to address dual-use AI,' he says. 'We can expect more international frameworks focused on responsible AI use in cyber defence, limits on offensive AI capabilities, and cross-border incident response cooperation. At QuantumGate, we've designed our products to support this shift and facilitate compliance with the country's cryptography regulations.'
Express Tribune
24-05-2025
- Business
- Express Tribune
Pakistan among least affected by web threats
Listen to article At its annual Cyber Security Weekend for the Middle East, Turkiye and Africa (META) region, the Kaspersky Global Research and Analysis Team presented cybersecurity trends, including ransomware, advanced persistent threats (APTs), supply chain attacks, mobile threats, artificial intelligence and IoT developments. The first quarter of 2025 showed that Turkiye and Kenya had the highest number of users affected by web incidents (online threats). They were followed by Qatar, Nigeria and South Africa. Saudi Arabia had the lowest while Pakistan had the second lowest share of users attacked by web-born threats in the META region. Kaspersky experts constantly track highly sophisticated attacks. Specifically, they are monitoring 25 APT groups currently active in the META region, including well-known ones such as SideWinder, Origami Elephant and MuddyWater. The rise of creative exploits for mobile and the further development of techniques aimed at evading detection are among the trends Kaspersky is seeing in these targeted attacks. Ramsomware remains one of the most destructive cyberthreats. According to Kaspersky data, the share of users affected by ransomware attacks increased 0.02 percentage point to 0.44% from 2023 to 2024 globally. In the Middle East, the growth is 0.07 percentage point to 0.72%, in Africa, there was a 0.01-percentage-point growth to 0.41% while Turkiye has a zero-percentage-point growth to 0.46%. Attackers often don't distribute this type of malware on a mass scale, but prioritise high-value targets. In 2025, ransomware is expected to evolve by exploiting unconventional vulnerabilities. The proliferation of large language models (LLMs) tailored for cybercrime will further amplify the ransomware's reach and impact. "Ransomware is one of the most pressing cybersecurity threats facing organisations today, with attackers targeting businesses of all sizes and across every region, including META. Ransomware groups continue to evolve by adopting techniques, such as developing cross-platform ransomware, embedding self-propagation capabilities and even using zero-day vulnerabilities that were previously affordable only for APT actors," said Sergey Lozhkin, Head of META and APAC regions in the Global Research and Analysis Team at Kaspersky.
Business Recorder
24-05-2025
- Business
- Business Recorder
Pakistan has second lowest share of users attacked by web-borne threats
ISLAMABAD: A global cybersecurity firm disclosed Saturday that Pakistan has the second lowest share of users attacked by web-borne threats during first quarter of 2025. In a new report-2025, Kaspersky Global Research and Analysis Team presented cybersecurity trends, including ransomware, advanced persistent threats (APTs), supply chain attacks, mobile threats, AI and IoT developments. First quarter of 2025 showed that Turkiye and Kenya had the highest number of users affected by web incidents (online threats). They were followed by Qatar, Nigeria and South Africa. Saudi Arabia had the lowest while Pakistan had the second lowest share of users attacked by web born threats in the META region during first quarter of 2025. Kaspersky experts constantly track highly sophisticated attacks. Specifically, they are monitoring 25 APT groups currently active in the META region, including such well-known ones as SideWinder, Origami Elephant, MuddyWater. The rise of creative exploits for mobile and further development of techniques aimed at evading detection among the trends Kaspersky is seeing in these targeted attacks. Ramsomware remains one of the most destructive cyberthreats. According to Kaspersky data, the share of users affected by ransomware attacks increased by 0.02 p.p to 0.44% from 2023 to 2024 globally. In the Middle East the growth is 0.07 p.p. to 0.72%, in Africa: 0.01 p.p. growth to 0.41%, in Turkiye 0.06 p.p. growth to 0.46%. Attackers often don't distribute this type of malware on a mass scale, but prioritize high-value targets. 'Ransomware is one of the most pressing cybersecurity threats facing organizations today, with attackers targeting businesses of all sizes and across every region, including META, said Sergey Lozhkin, Head of META and APAC regions in Global Research and Analysis Team at Kaspersky. 'To stay secure, organizations need a layered defense: up-to-date systems, network segmentation, real-time monitoring, robust backups, and continuous user education.' Copyright Business Recorder, 2025
