Over the past year, cybercriminals have increasingly targeted AI-based coding tools to access sensitive credentials, compromise systems, and breach data. According to a recent threat intelligence study, there were 103 separate incidents between July 31, 2025 and July 31, 2026 in which AI was either targeted, used, or abused to facilitate cyberattacks. Rather than fully autonomous attacks, the research indicates a trend of attackers integrating AI into existing techniques—such as phishing, supply-chain compromises, credential theft, and extortion—to gain leverage.
How Attackers Are Using Trusted Coding Tools
Tools designed to assist developers—coding assistants, plugins, dependencies—hold deep access into source code repositories, business systems, and credentials. When a dependency or plugin is compromised, attackers can piggyback on that trust to steal information, maintain persistence, and move laterally across corporate networks.
In one case from February 2026, a widely used AI proxy package that links various applications to large language models was hijacked. A malicious update enabled backdoor installation, credential siphoning, and horizontal access escalation. Another case in April involved AI-tool packages capturing developer credentials and penetrating crypto-wallets. Malicious plugins have also repurposed provider API keys supplied by users under false pretexts—redirecting sensitive credentials to attacker-controlled infrastructure.
The Scope and Nature of AI-Linked Incidents
The study classified 32 of the 103 incidents as directly involving AI in malware or hostile operations; the remaining 71 were attacks aimed at AI infrastructure itself or used AI as part of the delivery or deception. State-backed actors played a role in several of the more severe threats, including efforts that used AI to obfuscate code or to support intelligence gathering. One notable operation against government targets in Taiwan reportedly employed up to eight AI agents with minimal human oversight.
Despite often dramatic headlines, truly autonomous campaigns remain rare. AI is often a force multiplier rather than a replacement—used for crafting extortion messages, analyzing stolen data, or bypassing defenses. Successful attacks frequently depend on human-led orchestration. The report underscores that defenders need to monitor not only passwords but also access tokens, API keys, model repositories, and all tool dependencies.
To guard against breaches connected to AI tools, the report recommends stricter access controls—separating AI tools from critical systems—and enforcing approval workflows for sensitive actions. Regular audits of plugins, model packages, and dependencies are also advised, along with vigilant detection of leaked API credentials, stolen OAuth tokens, and misused AI subscriptions. Monitoring everything helps close off attack vectors before they’re exploited.
What this means: This wave of incidents serves as an urgent warning that the weakest links in AI environments may be the trusted tools developers already use daily. AI isn’t yet replacing human hackers, but it’s empowering them. Organizations that ignore dependency hygiene, access boundaries, or the risk from third-party plugins do so at their peril. Moving forward, watch which AI platforms and services reveal vulnerabilities—and which defenders adapt their toolchains before suffering breaches themselves.
