Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Microsoft Copilot Cowork Exfiltrates Files

promptarmor.com

Post reply on HN
Print view View original

31–40 of 58 posts

Re: Microsoft Copilot Cowork Exfiltrates Files

#31

Post by mlacks »

codebje wrote: post #28
mlacks wrote: post #25

Exfiltrates: to steal sensitive data from a computer system (for example, via a flash drive). I'm not going to defend Microsoft here, but the title (at the source blog) is misleading and a bit rage-baity. What happened with Cowork may have been rushed, possibly due to incompetence, but incompetence is not malice. This framing is also recycled across a few of the author's other interesting findings. Within the article…

The malice is by the author of the malicious skill file. This is an intrinsic risk associated with giving LLMs access to sensitive material. It's reckless of Microsoft to give an LLM such broad access based on the user's own permissions. If there were a confirmation prompt for the Teams message, why would even a highly competent user refuse it? That's what the skill says it will do. The message is expected, the visib…

agreed. Its the author of the skill file not literally Cowork

Re: Microsoft Copilot Cowork Exfiltrates Files

#32

Post by jychang »

mdavidn wrote: post #20
arjie wrote: post #12

A skill is just a program for an LLM agent. This just seems like works-as-expected. Are the five lines in the skill notably innocuous or something? I don't mean to dismiss it out of hand but I don't understand what happened here because it seems to read "`curl $url | bash` can exfiltrate data" which seems pretty straightforward that it can.

A skill is just instructions that the agent can autonomously copy into context. There’s no trust boundary between trusted and untrusted context.

Yeah, this is your fault if you install the skill.

This reads to me as "user installed exe file can upload your data to a server". Um, yes, that's the point?

This seems like this generation's equivalent of "don't open Linkin-Park.mp3.exe from limewire"

Re: Microsoft Copilot Cowork Exfiltrates Files

#33

Post by EFLKumo »

It's not the first time we hear about prompt injection attacks, and for sure it's the fault of Microsoft. Many talking about the prompt injection itself, whether Copilot should be able to defense prompt injections, etc. But that's not the problem.

OpenAI released their LLM-driven browser Atlas last year. Though their team is brilliant (https://openai.com/index/hardening-atlas-against-prompt-inje...), there has been a number of succeeded injection attacks.

IMO the real vulnerability is located at the "Act" part of "ReAct" (reasoning and action) agent framework.

> “[Copilot] Cowork asks for your permission before taking sensitive actions...” ... when the recipient is the active user, these actions execute immediately without requiring human approval (users do not have a setting to modify this behavior).

> Copilot Cowork can retrieve ‘pre-authenticated download links’ for files the user has access to, which allow anyone who opens the link to download that file.

> Microsoft Copilot Cowork has read access to essentially any resource a user does through Microsoft Graph. As such, the primary mechanism to reduce the blast radius of attacks like this is to restrict excessive permissioning across one’s Microsoft ecosystem.

Take it easy. Inside the whole attack flow, Microsoft gives Cowork unrestricted access and the ability to bypass approvals. I don't find much problem with LLMs here. It's said the attack is also a threat for Opus 4.7, but I've found several times Opus 4.7 forbidding context7.com's "prompt injections" only requiring opus to ask me creating an context7 API key to get more requests for free. From my personal experience, such models indeed are trained to perceive injections, but these injections could mask themselves as sth like Agent Skills, and there are always ways to win as red teams.

We may not lay our hope too much on defense of injections, but concentrating on restricting LLM's permissions. The popular usage of CLIs in agents' (especially coding agents) workflow has also concerned me since most cli tools an agent can access actually have the same permissions with users.

Re: Microsoft Copilot Cowork Exfiltrates Files

#34

Post by prpl »

arjie wrote: post #12

A skill is just a program for an LLM agent. This just seems like works-as-expected. Are the five lines in the skill notably innocuous or something? I don't mean to dismiss it out of hand but I don't understand what happened here because it seems to read "`curl $url | bash` can exfiltrate data" which seems pretty straightforward that it can.

Right, people haven’t internalized that these are really just scripts in natural language.

Re: Microsoft Copilot Cowork Exfiltrates Files

#36

Post by stingraycharles »

EFLKumo wrote: post #33

It's not the first time we hear about prompt injection attacks, and for sure it's the fault of Microsoft. Many talking about the prompt injection itself, whether Copilot should be able to defense prompt injections, etc. But that's not the problem. OpenAI released their LLM-driven browser Atlas last year. Though their team is brilliant ( https://openai.com/index/hardening-atlas-against-prompt-inje... ), there has been…

“IMO the real vulnerability is located at the "Act" part of "ReAct" (reasoning and action) agent framework.”

This is a fancy way of saying that “the problem is tool calling”, which is obviously true. The problem is that, when it works correctly (99.99% of the time), it adds so much more value to LLMs.

Sandboxing is a step in the right direction, but can also add friction.

Using guardrails is also good, but adds latency, expenses, and also doesn’t solve 100% of the issues.

IMHO there currently does not exist a proper solution to this problem, and it has yet to be discovered. The proper solution, however, should NOT be based on LLMs, so guardrails are the incorrect direction (albeit effective and easier to implement).

Re: Microsoft Copilot Cowork Exfiltrates Files

#37

Post by Forgeties79 »

stingraycharles wrote: post #36
EFLKumo wrote: post #33

It's not the first time we hear about prompt injection attacks, and for sure it's the fault of Microsoft. Many talking about the prompt injection itself, whether Copilot should be able to defense prompt injections, etc. But that's not the problem. OpenAI released their LLM-driven browser Atlas last year. Though their team is brilliant ( https://openai.com/index/hardening-atlas-against-prompt-inje... ), there has been…

“IMO the real vulnerability is located at the "Act" part of "ReAct" (reasoning and action) agent framework.” This is a fancy way of saying that “the problem is tool calling”, which is obviously true. The problem is that, when it works correctly (99.99% of the time), it adds so much more value to LLMs. Sandboxing is a step in the right direction, but can also add friction. Using guardrails is also good, but adds laten…

Ultimately it all sounds like variations of “don’t blame the tool for situations the tool enables,” which has never been particularly convincing as an argument if you ask me.

Re: Microsoft Copilot Cowork Exfiltrates Files

#38

Post by brookst »

pwarner wrote: post #15

MS rushed this to production, sure they call it a beta feature but it's clear it was super rushed. They're desperate to be relevant.

Bingo. MS has so many strengths that should make them relevant — a billion or so Windows installations, ~~Office~~ ~~M365~~ M365 Copilot is still the de facto productivity suite, Azure data centers, the OpenAI deal — and they just can’t get out of their own way because their strategy is “leverage those strengths to cram Copilot down peoples’ throats”

They have no taste. And no aspiration to taste. The industry is moving too quickly for the quasi-monopoly strategy of forcing users to buy their product.

Books will be written about how Microsoft had an amazing strategic position and failed at AI because they never prioritized an actual great product.

Re: Microsoft Copilot Cowork Exfiltrates Files

#39

Post by Yokohiii »

2001zhaozhao wrote: post #2

AKA, if a malicious skill got into your AI agent, you're cooked. I think this isn't surprising, nor do I think it should be considered a prompt injection at all. An AI skill is akin to a plugin for traditional software - if you install a malicious IDE extension or Outlook plugin, the attacker can also do whatever they want to the PC and exfiltrate whatever data they want to. So this article is a big nothingburger.

> nor do I think it should be considered a prompt injection at all

Can we stop the apologetic framing? It's increasingly common to create exploits from multiple vulnerabilities. Each one is bad. Downloading corporate malware is stupid. Adding random prompt injection is reckless. Insane to run autonomous agents on top of it.

Prompt injection is more serious in this regard, because there is no known solid protection. All the other problems are failure in process, prompt injection is failure at the first thought.

Re: Microsoft Copilot Cowork Exfiltrates Files

#40

Post by EFLKumo »

stingraycharles wrote: post #36
EFLKumo wrote: post #33

It's not the first time we hear about prompt injection attacks, and for sure it's the fault of Microsoft. Many talking about the prompt injection itself, whether Copilot should be able to defense prompt injections, etc. But that's not the problem. OpenAI released their LLM-driven browser Atlas last year. Though their team is brilliant ( https://openai.com/index/hardening-atlas-against-prompt-inje... ), there has been…

“IMO the real vulnerability is located at the "Act" part of "ReAct" (reasoning and action) agent framework.” This is a fancy way of saying that “the problem is tool calling”, which is obviously true. The problem is that, when it works correctly (99.99% of the time), it adds so much more value to LLMs. Sandboxing is a step in the right direction, but can also add friction. Using guardrails is also good, but adds laten…

By using "ReAct", I just wanted to emphasize the "agentic" perspective of tool calling, which makes tool calling facing the real world and at risk sometimes. So I'm not downplaying the significance of tool callings.

Yes I'm a builder of an agent infra on PCs, so I can completely sense that the protective measures are weak and inadequate, sometimes seeming like an unsolvable problem. But according to the article, what Microsoft did was hard to tell in a polite way. If they had even a little security awareness, I could completely understand, but it's like they've vibe coded the entire permissions system of Cowork.

Post reply on HN