The Danger of Blind Trust in AI Agents and the Threat to Software Supply Chains
Amid the rapid development of Artificial Intelligence (AI) that promises increased productivity, a new threat is emerging in the corporate world. Researchers from Kaspersky have highlighted the phenomenon of ‘blind trust’ in AI technology as a security gap that cybercriminals are now beginning to exploit.
Sojun Ryu, a security researcher at Kaspersky’s Global Research and Analysis Team (GReAT), explained that the evolution of AI has reached an autonomous stage. From being merely a content creation tool and copilot, the world is now entering the era of AI agents capable of planning, calling APIs, and acting independently. However, the speed of this adoption often sacrifices verification.
‘When speed outpaces verification, that speed can actually amplify risk,’ Ryu warned.
According to him, attackers are no longer focusing solely on technical vulnerabilities, but are instead exploiting users’ trust in popular AI services.
Kaspersky GReAT researchers found startling facts about tens of thousands of attacks disguised as well-known AI services. Attackers distributed fake versions of official applications to target users and developers.
In addition, more than 15,000 malware samples were found disguised as agentic AI software. This malware includes trojans, spyware, and backdoors that allow attackers to steal internal information and establish remote command-and-control access.
Ryu also emphasised that AI’s dependence on open-source packages is a significant weak point. Supply chain attacks on ecosystems such as npm and PyPI continue to increase in frequency.
One notable incident occurred in March 2026 involving Axios, a popular JavaScript library. After breaching the computer of the primary maintainer, attackers released a malicious version that was downloaded by hundreds of devices within just three hours. Given that Axios is used by more than 170,000 software packages, the impact of a single breach can spread widely and very quickly.
To maintain a balance between efficiency and security, Kaspersky recommends several practical steps for companies.
Ryu concluded that the security focus must shift to the early stages of development.
‘Security will fail if we prioritise speed over protection and visibility. When trust is verified before execution, organisations will actually move faster,’ he concluded.
Source: jawawa.id (https://jawawa.id/newsitem/the-danger-of-blind-trust-in-ai-agents-and-the-threat-to-software-supply-chains-1786949430)
Originally published by Media Indonesia (https://mediaindonesia.com/teknologi/922567/bahaya-kepercayaan-buta-pada-agen-ai-dan-ancaman-rantai-pasok-perangkat-lunak), translated from Indonesian