Skip to content
All articles
AI security

Prompt injection: the attack that needs no click

Conventional software keeps instructions and data apart, and a document cannot issue an order. A language model has no such boundary: everything it receives arrives as text, and text that looks like an instruction can be followed like one.

BastionUpdated 3 min read

OWASP ranks prompt injection first in its 2025 Top 10 for LLM applications, as LLM01. The list also covers the consequences: sensitive information disclosure at LLM02, excessive agency at LLM06, system prompt leakage at LLM07, and vector and embedding weaknesses at LLM08.

EchoLeak, the case that made it concrete

On 12 June 2025 The Hacker News reported a "zero-click" vulnerability that allowed attackers to “exfiltrate sensitive data from Microsoft 365 (M365) Copilot's context sans any user interaction”. It was found by Aim Security, given the identifier CVE-2025-32711, and scored 9.3.

The academic write-up describes it in one line: “EchoLeak (CVE-2025-32711), a zero-click prompt injection vulnerability in Microsoft 365 Copilot that enabled remote, unauthenticated data exfiltration via a single crafted email.”

"Zero-click" is literal: the victim did not open an attachment or approve anything. The assistant read the mail as part of its ordinary work, and the instruction hidden in it was executed against the data the assistant could already reach.

Microsoft fixed it server-side in May 2025, with no customer action needed, and listed the CVE in its June 2025 Patch Tuesday. The Hacker News reported that “There is no evidence that the shortcoming was exploited maliciously in the wild.”

The product was not unusually weak. The attack surface exists wherever an assistant reads untrusted text and can reach trusted data. Retrieval-augmented generation widens it, because every indexed document is model input, and an indexed document can be written by someone outside the company: a supplier's invoice, an emailed contract draft, a CV.

Running the model on-premise does not stop prompt injection

Sellers of local AI often imply that it does. Local inference stops a document being sent to an external service in normal operation. It does not stop a malicious document from steering the model: a local model that can read the shared drive and send email has the same structural problem as a hosted one.

What local deployment does change is the exfiltration path. An injected instruction still needs a way out, and EchoLeak's way out was an outbound fetch. A system with no outbound connectivity gives that instruction fewer places to send anything.

What reduces the risk

  • Keep retrieved content and instructions structurally separate, and treat every document as untrusted
  • Give the model the narrowest tool permissions that still do the job
  • Enforce the user's own access rights at retrieval time, not after generation
  • Validate outputs before another system acts on them
  • Constrain where the system may connect, so a successful injection has nowhere to send data
  • Test adversarially, with documents written to attack your own assistant

The rule under all of them is to give an AI system no more authority than the task requires. A model that reads documents, sends mail, queries the finance system and runs code is a different category of problem from a model that only reads documents.

With Bastion

What Bastion changes

Bastion is a private AI system delivered as one sealed box that runs inside your building. One monthly fee covers the hardware, the model, the hardened operating system and support, and nothing your team types leaves the building.

Bastion does not claim immunity from prompt injection, because no appliance can. What it closes is the easy exit: the unit never opens a connection to the internet on its own. How answers are rendered, and any tools, still need limits, though, because the user's browser can load a link planted in an answer.

The permissions, the retrieval rules and the tool access above that are still design decisions.

Questions this article answers

What is a prompt injection attack?
Text that a model reads as an instruction instead of as content: because a language model receives everything as text, a document, a web page or an email can carry an order the model then follows. OWASP ranks it first in its 2025 Top 10 for LLM applications, as LLM01.
What was EchoLeak, CVE-2025-32711?
A zero-click prompt injection in Microsoft 365 Copilot, disclosed by Aim Security in June 2025 and scored 9.3, in which one crafted email made the assistant exfiltrate data with no user interaction. Microsoft fixed it server-side in May 2025, and The Hacker News reported no evidence of exploitation in the wild.
Does running AI on-premise stop prompt injection?
No: local inference protects the data boundary, not the boundary between instruction and content. What it closes is the easy exit, because an injected instruction still needs somewhere to send data, but how answers are rendered, and any tools, still need limits, because the user's browser can load a link planted in an answer.

On the record

  1. 1
  2. 2
  3. 3
  4. 4