{"id":10181,"date":"2026-10-28T08:25:00","date_gmt":"2026-10-28T08:25:00","guid":{"rendered":"https:\/\/cybersecurityinfocus.com\/?p=10181"},"modified":"2026-10-28T08:25:00","modified_gmt":"2026-10-28T08:25:00","slug":"10-ways-attackers-abuse-ai-services-to-hack-your-business","status":"publish","type":"post","link":"https:\/\/cybersecurityinfocus.com\/?p=10181","title":{"rendered":"10 ways attackers abuse AI services to hack your business"},"content":{"rendered":"
\n
\n
\n
\n
<\/div>\n

Attackers are starting to exploit AI systems to mount attacks in the same way they once relied on built-in enterprise tools such as PowerShell.<\/p>\n

Instead of relying on malware, cybercriminals are increasingly abusing AI tools enterprises depend on \u2014 a trend some experts describe as \u201cliving off the AI land.\u201d The expanded abuse of AI ecosystems features a growing array of tactics, including poisoned models, compromised MCP servers, prompt injection attacks, and AI supply-chain vulnerabilities.<\/p>\n

\u201cWe\u2019re seeing it in things like poisoned MCP servers in the supply chain, attackers using legitimate models like Claude to extract sensitive data, and even viral agents like OpenClaw accidentally causing destructive actions,\u201d says Kaushik Shanadi, CTO at Helmet Security, a startup focused on securing agentic AI communications. \u201cThe problem is most of these systems were deployed before anyone stopped to think about governance or security.\u201d<\/p>\n

Agentic AI systems introduce new vectors of attack through delegated permissions, autonomous decision-making, and machine identities.<\/p>\n

The shift from simple prompt injection to \u201cagent hijacking\u201d represents a fundamental change in the AI threat landscape, other security experts tell CSO.<\/p>\n

\u201cAttackers are no longer just trying to trick a chatbot; they are living off the AI land, abusing the same legitimate automation and memory features that make AI assistants useful,\u201d says Pascal Geenens, VP of cyber threat intelligence at cybersecurity vendor Radware.<\/p>\n

Below are some examples of how attackers are subverting AI-based services to stage attacks.<\/p>\n

<\/a>MCP server impersonation<\/h2>\n

In September 2025, attackers promoted a counterfeit model context protocol (MCP) server mimicking technology to integrate Postmark<\/a>, a transactional email service owned by ActiveCampaign, into AI assistants.<\/p>\n

The fake MCP server package looked legitimate and functioned as a legitimate tool across 15 versions before a single-line code change was introduced that meant sensitive communications \u2014 password resets, invoices, internal memos \u2014 were silently siphoned off for days before the hack was detected.<\/p>\n

The malicious package, which attracted 1,500 downloads per week on the popular node.js package registry, exposed enterprises that relied on the tool to a form of supply chain attack.<\/p>\n

\u201cThis is the AI equivalent of name-squatting a package registry, except there\u2019s no central MCP authority verifying server identity and no cryptographic link between an MCP server and the organization it claims to represent,\u201d says Brad Micklea, CEO at Jozu, an AI security and MLOps platform. \u201cThis breaks the trust model before the MCP is deployed.\u201d<\/p>\n

MCP servers \u2014 which allow AI agents and chatbots to connect to data sources, tools, and other services<\/a> \u2014 have recently become the target of varied (for example against Cursor\u2019s built-in browser<\/a>) and sustained malicious attacks<\/a>. Locking down these systems<\/a> to minimize risks has become a priority for enterprise CISOs.<\/p>\n

\u201cThese servers expose tools, memory, and APIs to AI agents so they can perform tasks,\u201d says Zahra Timsah, PhD, CEO of i-GENTIC AI, an agentic AI governance platform. \u201cIf an attacker inserts a poisoned tool, modified connector, or malicious retrieval source into that chain, the AI agent can unknowingly execute it.\u201d<\/p>\n

<\/a>Abusing AI platforms as covert C2 channels<\/h2>\n

Cybercriminals are also abusing AI platforms as covert command-and-control (C2) channels by turning AI services into proxies that hide malicious traffic inside the flow of legitimate content.<\/p>\n

Instead of running a dedicated C2 server, malware is programmed to fetch commands and exfiltrate data through AI services, circumventing traditional security controls in the process.<\/p>\n

For example, the SesameOp backdoor<\/a> hid command traffic inside the OpenAI Assistants API, camouflaging instructions to malware as normal AI development activity.<\/p>\n

This is far from an isolated example and the potential for misuse is rife.<\/p>\n

For example, Check Point Research demonstrated how Microsoft Copilot and Grok might be manipulated<\/a> through their public web interfaces to fetch attacker-controlled URLs and return responses. This behavior opens the door to abuse of AI systems without requiring an API key or authenticated account.<\/p>\n

<\/a>Dependency poisoning in AI workflows<\/h2>\n

Rather than attacking an AI system directly some assaults have relied on poisoning downstream dependencies that an agent relies on for data processing.<\/p>\n

In one case, a compromised NPM package was injected into an agentic workflow\u2019s dependency chain<\/a>.<\/p>\n

\u201cThis mirrors classical supply chain attacks (e.g. SolarWinds), but a poisoned dependency in an agentic pipeline doesn\u2019t just leak data \u2014 it can alter the agent\u2019s decision-making, tool selection, or output without any visible anomaly,\u201d says Jozu\u2019s Micklea.<\/p>\n

Compromised npm packages were used to seed the spread of Shai-Hulud<\/a>, a self-propagating software supply-chain worm that first surfaced in September 2025.<\/p>\n

Shai-Hulud was designed to compromise developer credentials, steals secrets, and inject malicious install scripts into legitimate packages. These booby-trapped versions of reputable packages were published to target downstream developers and CI\/CD systems.<\/p>\n

ChainDrop targeted AI coding tool credentials<\/a> using a similar methodology in August 2026.<\/p>\n

<\/a>Double agents<\/h2>\n

Some attackers are weaponizing vulnerabilities in agents rather than abusing components of an enterprise\u2019s legacy IT infrastructure.<\/p>\n

For example, the \u201cEchoLeak\u201d command injection vulnerability in Microsoft 365 Copilot (CVE-2025-32711<\/a>) shows that a single email with concealed prompt-injection instructions is sufficient to force the AI assistant to exfiltrate internal files and emails to an external server without user interaction.<\/p>\n

A series of vulnerabilities (such as CVE-2026-25253<\/a>) in OpenClaw, the popular open-source personal AI assistant, created a route for a malicious website to take complete control of the developer\u2019s AI agent.<\/p>\n

\u201cMore than 21,000 such instances were detected, and the researchers further observed that 12% of the skills marketplace for the OpenClaw platform was distributing malware,\u201d says Dr. Suleyman Ozarslan, VP of Picus Labs at Picus Security, a specialist in breach and attack simulation.<\/p>\n

Security researchers at Varonis discovered an attack against Microsoft Copilot Personal<\/a> that sidestepped built-in AI safeguards simply by asking for sensitive data twice.<\/p>\n

The Reprompt vulnerability<\/a> \u2014 which effectively turned Microsoft Copilot into a data exfiltration tool \u2014 was reported to Microsoft, which has responded by issuing a patch.<\/p>\n

<\/a>AI-orchestrated espionage campaigns<\/h2>\n

Anthropic caught threat actors abusing Claude Code<\/a> to manage operational tasks in a cyber-espionage campaign in September 2025.<\/p>\n

A suspected Chinese state-sponsored group designated GTG-1002 used Claude Code to execute 80-90% of tactical operations independently<\/a>, at physically impossible request rates for human operators.<\/p>\n

Attackers abused the AI agentic capabilities of Claude Code to automate the process of scripting, target research, building attack tooling, and other functions.<\/p>\n

\u201cThe attackers decomposed their operation into thousands of small, individually innocuous tasks, combined with role-play framing that convinced the model it was operating as part of a legitimate security assessment,\u201d explains Yagub Rahimov, CEO at cybersecurity startup Polygraf AI.<\/p>\n

Agentic ransomware attacks<\/h2>\n

Miscreants abused an AI agent dubbed Jadepuffer<\/a> to run an end-to-end ransomware and extortion campaign in July 2026.<\/p>\n

Initial access was gained by exploiting a vulnerability in Langflow, an AI orchestration tool, before privilege elevation and the deployment of Encforge, ransomware that targets and encrypts roughly 180 AI-related file extensions. The attack ultimately targeted a victim\u2019s production database server.<\/p>\n

This wasn\u2019t the first fully autonomous cyberattack, but it was the first documented ransomware campaign run end-to-end by an agentic threat actor<\/a>, according to cloud security vendor Sysdig.<\/p>\n

\u201cIt destroyed a production database, wrote the ransom note, and went from failed login to working fix in 31 seconds,\u201d says Crystal Morin, senior cybersecurity strategist at Sysdig.<\/p>\n

Turning AI assets into attack targets<\/h2>\n

Threat actors are now prioritizing assaults against AI tools and infrastructure due to their high value.<\/p>\n

By compromising an AI infrastructure attackers gain access to LLM provider keys, cloud credentials, data, and myriad connections.<\/p>\n

\u201cTo get in, they [attackers] often scan for vulnerabilities and, as soon as one is live, use AI to turn advisories and disclosures into exploits,\u201d says Sysdig\u2019s Morin. \u201cIn many instances, we\u2019ve seen the mean time to exploit [MTTE] drop to just hours, like with the Marimo<\/a> and PraisonAI<\/a> CVEs.\u201d<\/p>\n

Attacking operational technology environments<\/h2>\n

Maintainers of industrial control systems have begun adopting AI technologies to increase the efficiency of their operations.<\/p>\n

Initially industrial AI was an advisor. If a model predicted a failure or suggested a better setpoint then it would still be up to a human operator to decide what to do. This is now changing with operators adopting more autonomous, agentic AI technologies.<\/p>\n

\u201cML models and AI agents now write setpoints back to control systems to optimize energy, yield and throughput, often every few minutes and without a human approving each change,\u201d says Thomas Plank, CEO of industrial cybersecurity specialists Tributech. \u201cProtocols such as MCP are accelerating this, because they give LLMs and agent frameworks a standard way into live OT [operational technology] and IoT [internet of things] data.\u201d<\/p>\n

These changes are opening up new attack surfaces.<\/p>\n

\u201cOnce AI acts on a physical process, attackers no longer need to compromise the controller, they can live off the AI\u2019s own authorized access,\u201d Plank warns.<\/p>\n

One attack scenario would involve poisoning the data the AI acts on so that decisions are made against manipulated sensor data rather than actual readings.<\/p>\n

\u201cIf values are tampered with, stale or simply wrong, the model makes a wrong decision with full confidence,\u201d Plank explains. \u201cAn attacker does not need to touch the AI at all; changing the data it reads, or the data it was trained on, is enough.\u201d<\/p>\n

Plank concludes: \u201cIn a closed loop, that wrong decision becomes a physical action.\u201d<\/p>\n

Using stolen compute to build offensive infrastructure<\/h2>\n

Threat actors have been stealing stolen AI computing resources since early 2024, a form of attack known by some as LLMjacking.<\/p>\n

Attackers began LLMjacking to resell access but more recently some have moved onto using stolen resources to build malicious tools.<\/p>\n

For example, in June 2026 the Sysdig Threat Research Team documented an operator using a victim\u2019s exposed, abliterated Ollama model to build an autonomous offensive framework<\/a>.<\/p>\n

<\/a>Creating modular black-hat AI platforms<\/h2>\n

The threat landscape has shifted from abusing chatbots to building dedicated, weaponized AI stacks like Xanthorox AI.<\/p>\n

Unlike general-purpose LLMs, Xanthorox is a purpose-built offensive platform designed specifically for cybercrime. The platform features modules for functions such as malware generation and vulnerability exploits.<\/p>\n

\u201cHexstrike AI MCP integration allows Xanthorox to move beyond mere \u2018assisted\u2019 hacking into the realm of fully autonomous agent systems, moving it into the realm of \u2018vibe hacking,\u2019\u201d says Radware\u2019s Geenens. Hexstrike is an open-source, AI-powered offensive security framework originally designed for ethical penetration testing.<\/p>\n

<\/a>Check against delivery<\/h2>\n

Zbyn\u011bk Sopuch, CTO of cybersecurity vendor Safetica, says that many attackers are no longer just exploiting software vulnerabilities, preferring instead to exploit the trust organizations place in AI.<\/p>\n

\u201cThis means security teams need to treat AI assistants the same exact way they treat human privileged users: with tight control, specific monitoring, and most importantly, never assume anyone or anything to be safe,\u201d Sopuch concludes.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>","protected":false},"excerpt":{"rendered":"

Attackers are starting to exploit AI systems to mount attacks in the same way they once relied on built-in enterprise tools such as PowerShell. Instead of relying on malware, cybercriminals are increasingly abusing AI tools enterprises depend on \u2014 a trend some experts describe as \u201cliving off the AI land.\u201d The expanded abuse of AI […]<\/p>\n","protected":false},"author":0,"featured_media":10182,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-10181","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-education"],"_links":{"self":[{"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=\/wp\/v2\/posts\/10181"}],"collection":[{"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=10181"}],"version-history":[{"count":0,"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=\/wp\/v2\/posts\/10181\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=\/wp\/v2\/media\/10182"}],"wp:attachment":[{"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=10181"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=10181"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cybersecurityinfocus.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=10181"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}