{"slug": "kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain", "title": "Kaspersky Warns Blind Trust in AI Agents Is Fueling a New Wave of Supply Chain Attacks", "summary": "Kaspersky's Global Research and Analysis Team (GReAT) warned that blind trust in AI agents is fueling a new wave of supply chain attacks, with 92,000 malicious attacks in 2026 disguised as legitimate AI services, 49% impersonating ChatGPT and 18% each posing as Claude and Gemini. The researchers also found over 15,000 malware samples disguised as agentic AI software, including trojans, spyware, and backdoors, and noted at least ten large-scale attack campaigns on open-source ecosystems like npm and PyPI since mid-2025. Security researcher Sojun Ryu said attackers exploit the gap where verification is skipped in favor of productivity.", "body_md": "AI agents are no longer just productivity tools sitting inside a company’s workflow. According to [Kaspersky](https://www.kobaran.com/tag/Kaspersky), they have quietly become a new layer of the software supply chain, and attackers have noticed long before most security teams have.\n\nThat warning came from Kaspersky’s Global Research and Analysis Team (GReAT) during the company’s Asia Pacific Cyber Security Weekend, held recently in Guangzhou, China. The message was blunt: as organizations race to adopt AI agents that can plan tasks, call APIs, and act with growing autonomy, the humans who once verified every step of that process are increasingly stepping back, and attackers are exploiting exactly that gap.\n\nThe timing matters. Enterprises across the region are under pressure to move faster with AI, often treating speed as the main measure of success. Kaspersky’s findings suggest that this same urgency is what cybercriminals are counting on, since it is far easier to abuse misplaced trust than to break through a technical defense.\n\n## Why Kaspersky Says Trust, Not Vulnerabilities, Is the New Attack Surface\n\nKaspersky GReAT security researcher Sojun Ryu framed the shift in AI adoption as a series of stages that happened faster than most security practices could keep up with. AI tools began as simple content generators, evolved into copilots embedded directly in daily workflows, and have now reached a point where autonomous agents can plan actions and execute tasks with minimal human oversight.\n\nRyu pointed out that even though people are technically still in the loop at each stage of AI-assisted development, verification is frequently skipped in the name of productivity. That gap, he explained, is where risk accumulates, because attackers do not need to find a software flaw when they can simply exploit the fact that a user or system trusts an AI tool without double-checking it.\n\n### The Scale of Fake AI Applications Targeting Users\n\nThe numbers Kaspersky shared illustrate how organized this exploitation has become. The company’s researchers tracked 92,000 malicious attacks in 2026 that were disguised as legitimate AI services. Roughly half of those, about 49 percent, impersonated ChatGPT applications, while fake versions posing as Claude and Gemini each accounted for 18 percent of the attacks.\n\nThese particular tools were targeted precisely because they sit at the entry point of most AI development environments. Attackers understand that developers and everyday users alike trust these names, so distributing convincing fake versions gives them a direct line into a target’s system.\n\n### Malware Hiding Inside Agentic AI Software\n\nBeyond fake applications, Kaspersky’s team also uncovered more than 15,000 malware samples disguised specifically as agentic AI software. These samples were not simple nuisances. Researchers classified them across several categories, including trojans, spyware, exploit kits, downloaders, droppers, and backdoors, each carrying real malicious functionality.\n\nIn practical terms, a single misplaced click on one of these disguised tools could be enough for an attacker to extract sensitive internal data or quietly establish long-term command-and-control access inside a company’s network.\n\n## Open Source Packages Remain a Prime Target\n\nRyu also emphasized a separate but closely related risk: the open-source ecosystems that both developers and AI systems depend on every day. He noted that open source is essentially unavoidable, since developers rely on it and AI models are trained and built using it, which makes it an attractive target for attackers looking to reach as many victims as possible through a single compromised package.\n\nAccording to Kaspersky, at least ten large-scale attack campaigns have hit open-source ecosystems, particularly npm and PyPI, since the middle of last year, with the pace of these incidents continuing to climb. The threats range from attacks on widely used packages to self-replicating worms capable of spreading without further human input.\n\n### The Axios Compromise as a Case Study\n\nOne incident stood out as a clear example of how much damage a single breach can cause. In March 2026, attackers compromised the machine of a lead maintainer behind Axios, one of the most widely used JavaScript libraries in the world. Axios is downloaded more than 100 million times every week and is used across over 170,000 software packages, making it a high-value target.\n\nOnce the attackers gained access to the maintainer’s npm account, they published tampered versions of the library. Those malicious versions stayed live for only about three hours before being caught, yet in that short window they were still downloaded by hundreds of devices. Kaspersky says the incident is a clear illustration of how quickly a single point of compromise in a trusted open-source project can ripple outward across the broader supply chain.\n\n### How Widespread the Problem Already Is\n\nKaspersky’s own enterprise research reinforces how significant this exposure has become. Company survey data indicates that 31 percent of enterprise businesses were affected by a supply chain attack over the past year, underscoring just how deeply open-source components are woven into modern development environments.\n\nBelow is a snapshot of the key figures Kaspersky shared during the event.\n\n| Metric | Figure |\n|---|---|\n| Malicious attacks disguised as AI services (2026) | 92,000 |\n| Share disguised as ChatGPT | 49% |\n| Share disguised as Claude | 18% |\n| Share disguised as Gemini | 18% |\n| Malware samples disguised as agentic AI software | 15,000+ |\n| Large-scale open-source attack campaigns since mid-last year | 10+ |\n| Enterprises affected by supply chain attacks | 31% |\n| Axios weekly downloads | 100 million+ |\n| Software packages relying on Axios | 170,000+ |\n| Misconfigurations found in audited CI/CD workflows | 250,000+ |\n\n## What Kaspersky Recommends to Reduce Risk Without Slowing Teams Down\n\nRecognizing that most organizations will not simply abandon AI tools or open-source software, Kaspersky’s guidance centers on tightening verification without adding friction that pushes teams to bypass security altogether.\n\n### Establishing Clear Trust Boundaries\n\nRyu recommended that organizations define explicit trusted development zones that separate external content from internal development assets. That includes hardening the integrated development environment itself, along with the extensions, workspaces, and permissions granted to AI agents operating within it.\n\n### Visibility Over More Approval Prompts\n\nRather than layering on additional approval steps that developers are likely to click through without reading, Kaspersky’s approach favors controlling the paths through which software actually enters an environment and maintaining continuous visibility into that activity. The goal, as Ryu described it, is to make the secure path the path of least resistance, not an obstacle.\n\n### Kaspersky’s Own Role in Monitoring the Ecosystem\n\nKaspersky said its GReAT team already monitors open-source software across major ecosystems and maintains a feed that flags vulnerable or malicious components as they emerge. The company’s researchers also used its Container Security capability to audit GitHub Actions workflows, a review that surfaced more than 250,000 potential misconfigurations in continuous integration and continuous delivery pipelines.\n\nThat figure, Kaspersky noted, points to how common insecure CI/CD configurations still are across the industry, even as organizations increasingly rely on automated pipelines to ship code faster.\n\n## The Bigger Picture for Enterprise Security Teams\n\nRyu’s closing point was less about a single fix and more about a shift in mindset. He argued that security needs to move earlier in the software development lifecycle and expand beyond protecting systems that are already deployed to securing the environments where software is actually built. When trust is verified before code executes, he said, organizations end up moving faster overall rather than slower, because they avoid the costly cleanup that follows a breach.\n\nFor enterprises weighing AI adoption against security exposure, Kaspersky’s research offers a clear takeaway: the tools themselves are not necessarily the weak point. The willingness to trust them without verification is.\n\nThis is a developing area of research, and Kaspersky has indicated that its GReAT team will continue tracking new campaigns targeting AI agents and open-source ecosystems as they emerge.", "url": "https://wpnews.pro/news/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain", "canonical_source": "https://www.kobaran.com/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain-attacks/", "published_at": "2026-08-24 03:37:00+00:00", "updated_at": "2026-08-24 03:42:50.729752+00:00", "lang": "en", "topics": ["ai-safety", "ai-policy", "ai-agents", "ai-infrastructure"], "entities": ["Kaspersky", "Global Research and Analysis Team (GReAT)", "Sojun Ryu", "ChatGPT", "Claude", "Gemini", "npm", "PyPI"], "alternates": {"html": "https://wpnews.pro/news/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain", "markdown": "https://wpnews.pro/news/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain.md", "text": "https://wpnews.pro/news/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain.txt", "jsonld": "https://wpnews.pro/news/kaspersky-warns-blind-trust-in-ai-agents-is-fueling-a-new-wave-of-supply-chain.jsonld"}}