Hackers find a way to exploit AI Agents through users’ Downloads folders | Find a Way

Hackers find a way to exploit AI agents through users' Downloads folders

Beware: Your Downloads folder could hold instructions for your AI assistant that you definitely didn’t write.
By on
Credit: Getty Images

Your Downloads folder probably contains a few PDFs, forgotten screenshots, and several files named some variation of "final." But according to a recent report by security company Forever, it could also contain instructions telling your AI assistant to install malware.

In an Oct. 22 blog post, the company reported that they caught an AI coding assistant following those instructions on a customer's computer. The company blocked the installation, then traced the request to a text file in Downloads.

How does a text file get that kind of power? Tools like OpenAI's Codex and Anthropic's Claude Code can, depending on their permissions, edit files, run commands, and install software. Developers leave them instruction files called AGENTS.md or CLAUDE.md to explain how to work on a project. If you're building a website, those notes might tell the assistant which fonts to use or what to test before finishing.

An attacker can leave instructions, too. In this case, the file told the assistant to install "function-color," a software package linked to the MALFEX malware campaign documented by Checkmarx.

SEE ALSO:PHP vulnerability: PHP security issues today explained

Forever didn’t establish how the file reached the customer's computer, but its researchers demonstrated one possible route. Depending on browser settings, a website can drop a file into Downloads without asking for approval. If someone then launches their coding assistant in that folder, it could read the file and follow its orders.

When Forever tested the customer's file on an isolated computer, it said Codex downloaded and ran the malware. It also changed settings that could make the attacker’s software load in later sessions.

"Once it is written, every Codex session on the machine loads the attacker’s server at startup — in any folder, any project," the researchers wrote. "You can leave the folder. The poison stays."

The team separately devised a test targeting Claude Code. The attack worked about half the time, redirecting Claude's traffic to a server the researchers controlled. That exposed a login token and allowed them to send commands back for the assistant to execute, illustrating the risk of giving an assistant access to your computer when it can mistake a stranger's instructions for its own assignment.

Researchers have pulled off similar tricks before, showing what AI can really do in the wrong hands. In March 2025, Pillar Security researchers got AI coding assistants to slip harmful code into a project by hiding malicious instructions in their coding rules. In April 2026, NVIDIA’s researchers used a poisoned AGENTS.md file to get Codex to sabotage software and conceal the change. This time, though, Forever says it caught an actual attacker trying the trick on a customer's computer. Yikes.

How to protect yourself

Forever recommends keeping coding assistants out of your Downloads folder and checking unfamiliar AGENTS.md or CLAUDE.md files before letting an agent work in a project. Keep sandboxes enabled to restrict what the assistant can access, and leave approval requirements on so the agent has to ask permission before taking certain actions.

If you suspect an assistant has already followed malicious instructions, check its settings for unfamiliar servers or connections and revoke and replace potentially exposed login tokens. Deleting the original text file may not undo changes the assistant has already made.

"The thing to watch is not this file or this folder," the researchers wrote. "It is the agent itself — software that reads untrusted input, holds your credentials, and acts with your authority, sitting in a blind spot the old tools were never built to see."

Want more science and tech news delivered straight to your inbox? Sign up for Mashable's Light Speed newsletter or get Mashable push alerts today.

Topics Cybersecurity


Matthews Martins

Perhaps facing reality head on is the most honest way to try to escape it.

93 Comments

Stay informed!

  1. downloads folder is basically a junk drawer of random pdfs, wild that agents just treat it like trusted input

    ReplyDelete
  2. Imagine getting hacked because of a random text file in your Downloads folder 💀

    ReplyDelete
  3. AI agents are only as safe as the permissions we give them. People need to understand that.

    ReplyDelete
  4. So now I have to be scared of my own Downloads folder? 😂

    ReplyDelete
  5. This is exactly why AI shouldn't be allowed to execute commands without asking the user first

    ReplyDelete
  6. The scary part is that the AI might think it's doing exactly what it's supposed to do

    ReplyDelete
  7. People keep saying AI will replace programmers, but now we also need programmers to protect us from AI.

    ReplyDelete
  8. Honestly, most people don't even check what they're downloading. This could become a serious problem.

    ReplyDelete
  9. Imagine an AI assistant accidentally handing over your credentials because it trusted a file. That's wild

    ReplyDelete
  10. Security researchers have been warning about prompt injection for a while. This shouldn't surprise anyone.

    ReplyDelete
  11. Convenience always comes with a price. Giving an AI access to your entire computer is a big risk

    ReplyDelete
  12. The real question is why the AI was able to install anything without explicit permission

    ReplyDelete
  13. First, we had to worry about suspicious links. Now we have to worry about suspicious instructions for our AI. 💀

    ReplyDelete
  14. I love AI tools, but I would never give them unrestricted access to my personal computer.

    ReplyDelete
  15. This sounds like a problem that better sandboxing and stricter permissions should help prevent.

    ReplyDelete
  16. Everyone wants autonomous AI until it starts autonomously doing things they never asked for

    ReplyDelete
  17. Wait until regular users start using coding agents without understanding what permissions they've enabled

    ReplyDelete
  18. The technology is impressive, but security needs to be part of the design from day one.

    ReplyDelete
  19. Not gonna lie, I would probably never have thought to check a random text file in my Downloads folder

    ReplyDelete
  20. This isn't just an AI problem. It's also a reminder that untrusted files can be dangerous when software automatically acts on their contents

    ReplyDelete
  21. AI is becoming more powerful, and the consequences of trusting it blindly are becoming more serious

    ReplyDelete
  22. Wow, so now we have to worry about our AI assistants getting tricked too? 😅

    ReplyDelete
  23. This is actually scary. Imagine getting hacked just because you downloaded a random file without even opening it

    ReplyDelete
  24. Never thought my Downloads folder could become a security risk like this. Crazy stuff.

    ReplyDelete
  25. This is why I never trust random files on my PC. Now AI agents need the same level of caution.

    ReplyDelete
  26. That’s wild. The AI is basically doing the hacker’s job for them without even realizing it.

    ReplyDelete
  27. ngl, the part about the attack staying active even after leaving the folder is pretty terrifying.

    ReplyDelete
  28. And here I was thinking AI assistants were supposed to make our lives easier lol.

    ReplyDelete
  29. Another day, another reason to be careful online. Stay safe out there, folks!

    ReplyDelete
  30. This highlights a serious security weakness in AI agents: the ability to execute untrusted instructions with the user's permissions.

    ReplyDelete
  31. The real concern here is the combination of broad system access, persistent changes, and insufficient verification of external instructions.

    ReplyDelete
  32. AI security needs to go beyond model safety. Permission management and isolation are just as important.

    ReplyDelete
  33. This is a good reminder that files containing instructions should be treated as untrusted input, regardless of their format or location.

    ReplyDelete
  34. The potential exposure of authentication tokens makes this much more concerning than a simple malicious file.

    ReplyDelete
  35. Sandboxing and explicit user approval should be fundamental safeguards, not optional features.

    ReplyDelete
  36. The problem isn't just that AI can make mistakes. It's that those mistakes can have real consequences when the system has access to your computer.

    ReplyDelete
  37. Convenience is great until your AI assistant starts following instructions from a complete stranger

    ReplyDelete
  38. We’re basically entering an era where social engineering attacks can target both humans and the AI tools they use.

    ReplyDelete
  39. I wonder how many users realize just how much access they’ve given these coding assistants.

    ReplyDelete
  40. This makes me question whether AI agents should automatically trust any instruction file they find on a system.

    ReplyDelete
  41. The technology is moving fast, but security practices really need to catch up.

    ReplyDelete
  42. Well… that's unsettling 😬

    ReplyDelete
  43. AI agents getting social-engineered now. What a time to be alive.

    ReplyDelete
  44. yikes, definitely turning off some permissions after reading this.

    ReplyDelete
  45. This is actually insane ngl.

    ReplyDelete
  46. So basically, don't blindly trust your AI. Got it.

    ReplyDelete
  47. bruh 💀 even my Downloads folder isn't safe anymore

    ReplyDelete
  48. Security is gonna get a LOT more complicated from here.

    ReplyDelete
  49. That's a big yikes. Good thing they caught it.

    ReplyDelete
  50. Well, that's scary 😬

    ReplyDelete
    Replies
    1. scary is what happening in world

      Delete
  51. Yikes, this is getting wild.

    ReplyDelete
  52. Never trusting my Downloads folder again 💀

    ReplyDelete
  53. That's actually crazy ngl

    ReplyDelete
  54. Hackers are getting smarter fr

    ReplyDelete
  55. Welp, time to check my files 😂

    ReplyDelete
  56. This is kinda terrifying tbh.

    ReplyDelete
  57. Bro, even AI can get hacked now?

    ReplyDelete
  58. Cybersecurity is getting wild these days.

    ReplyDelete
  59. Imagine getting hacked by a txt file 💀

    ReplyDelete
  60. That's messed up fr.

    ReplyDelete
  61. AI needs better security asap.

    ReplyDelete
  62. Scary stuff, stay safe everyone!

    ReplyDelete
  63. This is why permissions matter.

    ReplyDelete
  64. Another reason to be cautious online.

    ReplyDelete
  65. Didn't see that coming tbh.

    ReplyDelete
  66. Crazy how easily this can happen.

    ReplyDelete
  67. Wait, a simple text file?! 😳

    ReplyDelete
  68. That's lowkey terrifying.

    ReplyDelete
  69. Hackers never run out of ideas smh.

    ReplyDelete
  70. Definitely something to watch out for.

    ReplyDelete
  71. AI agents need serious guardrails.

    ReplyDelete
  72. Welp, that's enough internet for today 😂

    ReplyDelete
  73. This could get ugly real fast.

    ReplyDelete
  74. Kinda scary how much access AI has.

    ReplyDelete
  75. Fr, never underestimate these attacks.

    ReplyDelete
  76. Stay alert out there, folks!

    ReplyDelete
  77. That's actually a huge security risk.

    ReplyDelete
  78. Crazy times we're living in.

    ReplyDelete
  79. Damn, that's wild.

    ReplyDelete
  80. RIP to those who never check their files 💀

    ReplyDelete
  81. More security, less blind trust.

    ReplyDelete
  82. This is no joke tbh.

    ReplyDelete
  83. AI security is a whole new challenge.

    ReplyDelete
  84. And things are only getting crazier 😅

    ReplyDelete
  85. One file can cause all that? Wow.

    ReplyDelete
  86. Good reminder to double-check everything.

    ReplyDelete
Post a Comment
Previous Post Next Post