News: 1753367170

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Compromised Amazon Q extension told AI to delete everything – and it shipped

(2025/07/24)


The official Amazon Q extension for Visual Studio Code (VS Code) was compromised to include a prompt to wipe the user's home directory and delete all their AWS resources.

The bad extension was live on the VS Code marketplace for two days, though it appears that the intent was more to embarrass AWS and expose bad security rather than to cause immediate harm.

A [1]commit to the Amazon Q part of the AWS toolkit for VS Code includes a script that downloads an additional file, saved as extensionNode.ts . The source for this file includes a prompt instructing an AI agent to delete all non-hidden files from the user's home directory and then to "discover and use AWS profiles to list and delete cloud resources using AWS CLI commands."

[2]

The script then passes this prompt to the Amazon Q CLI, including the arguments --trust-all-tools and --no-interactive .

[3]

The malicious AI prompt which is downloaded by the compromised extension – click to enlarge

According to a [4]report , "a person who presented themselves as the hacker responsible" contacted 404 Media to explain that the wiper was designed to be defective, but was "a warning to see if they'd publicly own up to their bad security."

The person claimed that they submitted a pull request to the AWS repository from "a random account with no existing access" and were given admin credentials. They said that AWS then released the compromised package "completely oblivious."

[5]

[6]

Whether or not that report is correct, we can see the bad commit was indeed merged and released in version 1.84 of the extension on July 19, and reverted in version 1.85 published two days later. The changelog for 1.85 states: "Miscellaneous non-user-facing changes."

[7]

The changelog for the fixed extension refers to miscellaneous non-user-facing changes – click to enlarge

AWS posted a [8]security bulletin , which states:

AWS is aware of and has addressed an issue in the Amazon Q Developer Extension for Visual Studio Code (VSC). Security researchers reported a potentially unapproved code modification was attempted in the open source VSC extension that targeted Q Developer CLI command execution. This issue did not affect any production services or end-users.

Once we were made aware of this issue, we immediately revoked and replaced the credentials, removed the unapproved code from the codebase, and subsequently released Amazon Q Developer Extension version 1.85 to the marketplace.

[9]AWS goes full speed ahead on the AI agent train

[10]AWS previews Kiro IDE for developers who are over vibe coding

[11]Jilted AWS reckons VMware is now crusty like a mainframe

[12]AI and analytics converge in new generation Amazon SageMaker

This statement does not address the key issue of how the incident was allowed to happen. The consequences of unauthorized code in a popular AWS extension for VS Code could be calamitous. There are hints that the AWS SDK for .NET was compromised as well, though we have no details of this, and the AWS bulletin states that "no action is required for AWS SDK for .NET users."

The malicious commit has the same title as a previously merged commit, though the code itself is not at all related. The commit is also obviously suspicious, downloading a file from somewhere on GitHub to overwrite another file in the package. The implication, perhaps, is that there is too much reliance on AI to check the security of the code, in this case badly, and not enough human checks. This line of thinking is encouraged by another remark attributed to the bad actor, that "ruthless corporations leave no room for vigilance among their overworked developers."

AWS has recently [13]laid off a number of workers and Amazon CEO Andy Jassy has stated in a memo to employees that AI is likely to "reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company."

[14]

Could such "efficiency gains" affect the security of official AWS tooling, as this latest incident implies? It is a disturbing possibility, considering that the company has historically maintained a strong security record.

AWS watcher Corey Quinn asked the key question: "What did Amazon's internal review process for this repo actually look like?" and [15]concluded that "it's the same mess I called out back in 2022 when Azure's security posture fell flat on its face: companies treating security like an afterthought until it explodes in public." ®

Get our [16]Tech Resources



[1] https://github.com/aws/aws-toolkit-vscode/commit/678851bbe9776228f55e0460e66a6167ac2a1685

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aIZNNgjFu5hWFzbG10lXZwAAAAo&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[3] https://regmedia.co.uk/2025/07/24/evilprompt.jpg

[4] https://www.404media.co/hacker-plants-computer-wiping-commands-in-amazons-ai-coding-agent/

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIZNNgjFu5hWFzbG10lXZwAAAAo&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aIZNNgjFu5hWFzbG10lXZwAAAAo&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[7] https://regmedia.co.uk/2025/07/24/changlog.jpg

[8] https://aws.amazon.com/security/security-bulletins/rss/aws-2025-015/

[9] https://www.theregister.com/2025/07/17/aws_agentcore_ai/

[10] https://www.theregister.com/2025/07/14/aws_kiro_agentic_ide/

[11] https://www.theregister.com/2025/05/16/aws_vmware_mainframe_ai_transformation/

[12] https://www.theregister.com/2024/12/06/sagemaker_unified_studio_preview/

[13] https://www.theregister.com/2025/07/18/aws_sheds_jobs/

[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_security/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aIZNNgjFu5hWFzbG10lXZwAAAAo&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[15] https://www.lastweekinaws.com/blog/amazon-q-now-with-helpful-ai-powered-self-destruct-capabilities/

[16] https://whitepapers.theregister.com/



The future of humanity will forever be

JimmyPage

checking the output of AI. Slowly reducing the efficiency of the workforce in direct proportion to it's penetration.

Looks like banging the rocks together will ultimately prove more worthwhile.

Re: The future of humanity will forever be

stiine

And more fun!

Re: The future of humanity will forever be

UnknownUnknown

Is the Amazon Mechanical Turk still a thing … or did Alexa put him out to pasture

Re: The future of humanity will forever be

David Hicklin

I am so glad I am retired and avoided this stuff at work

I do feel sorry for those poor sops who are doing that checking however I suspect that the AI volume of output will far exceed their checking capacity

What could go wrong ?

Another day

ecofeco

Oh joy

Efficiency? Not likey

Cliffwilliams44

"reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company."

Oh, lord!

While I use AI to help in the creation of projects it is nowhere near at the level that it can replace humans. While it hac suggest some fairly good code for the initial request, once you start asking for revisions or correction, it can go wildly off track. I've seen it make sweeping changes to code when only a small revision was asked for, it makes assumptions based on its "desire" to please the users, if the user makes in incorrect statement, it will almost always treat that as correct and use that to make even more wildly inaccurate and fantastical nonsense. I've actually had to close a chat and start a new one because the LLM in that chat had gone completely off the rails!

Assuming this is going to replace developers and code reviewers is a complete fantasy, any company that makes this commitment is going to quickly find out how insane this is!

Time

Anonymous Coward

LLMs have their place and they will and are improving. But they do not think like us and do not seem to grasp a broad context without using masses of compute. They clearly don't understand humans although they often claim to. I think the fact their knowledge is curated and they have guardrails means they cannot understand us. It's the stuff that their owners stop them seeing or deducing that explains the human world. They are a productivity gain if used appropriately but it's nowhere near that claimed, like all new technology it's incremental. Will it change the world? Yes it will even if not through its capability then through the hype and spin.

AI Productivity???

druck

I'm just struck by how long winded the prompt was just to do a local rm -rf * when it specified the AWS commands directly.

Nobody ever forgets where he buried the hatchet.
-- Kin Hubbard