Attackers abuse enterprise AI tools for theft, C2, and agent hijacking
Original title: 10 ways attackers abuse AI services to hack your business
A fake Postmark MCP package siphoned mail after 15 clean versions, and SesameOp hid C2 in the OpenAI Assistants API.
CSO reports criminals are “living off the AI land,” abusing models, MCP servers, and agent permissions instead of traditional malware. Cited cases include a counterfeit Postmark MCP server, SesameOp command traffic in the OpenAI Assistants API, npm poisoning tied to Shai-Hulud, EchoLeak in Microsoft 365 Copilot, and flaws in OpenClaw. Experts say many agent systems were deployed before governance or identity controls existed.
Unconfirmed: The piece is an expert roundup; it does not independently confirm current exploitation of every cited flaw.
- The fake Postmark MCP package drew about 1,500 weekly npm downloads before a one-line change.
- EchoLeak, CVE-2025-32711, can exfiltrate Microsoft 365 files from one email without user interaction.
- CVE-2026-25253 and related OpenClaw flaws were linked to more than 21,000 exposed instances.
- Shai-Hulud, seen in September 2025, stole developer secrets and poisoned npm install scripts.
- Check Point showed Copilot and Grok web interfaces could fetch attacker-controlled URLs without an API key.
Impact: Data theft, covert C2, agent takeover, and supply-chain compromise
| Product | Affected |
|---|---|
| Microsoft 365 Copilot | — |
| OpenClaw | — |
| ActiveCampaign Postmark MCP integration | — |
- 1.Treat untrusted MCP servers and npm dependencies as supply-chain risk before connecting them to agents.
- Malware
- SesameOpShai-HuludChainDrop
- Organizations
- RadwareCheck PointJozuHelmet Security
Vulnerabilities mentioned
All| CVE | Vulnerability | CVSS | EPSS | Signals | Published |
|---|---|---|---|---|---|
| CVE-2025-32711 | Ai command injection in M365 Copilot allows an unauthorized attacker to disclose information over a network | 7.5 | 7% | ||
| CVE-2026-25253 | One-Click WebSocket Token Leak Enabling RCE in OpenClaw (Moltbot/Clawdbot) | 8.8 | 32% | PoC ×4 |
From the original article
7 min read at csoonline.com
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 — a trend some experts describe as “living off the AI land.”
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.