“Why are these systems even on a network? Just disconnect them!”
I see this after every OT attack, usually from smart IT people. As someone who responds to these incidents, I wish it were that easy. I wrote up how we got here, and why water utilities and refrigerators keep getting hit.
https://tisiphone.net/2026/11/08/lets-talk-about-why-all-this-ot-stuff-is-connected/
Cyberplace.social Mastodon server is 4 years old this month :party_porg:
It's the 6th most popular server on the covenant list with open registrations.
During the time it's been live, 24805 other fediverse servers have stopped responding.
In fediverse time we're about 100 years old.
The portable aircon units in the picture were the aircon at the core datacenter of a billion pound multinational I worked at.
Yes guys, the answer is capitalism.
With a side of convenience and laziness capitalism.
But we're stuck with the problem.
It is ridiculously easy to backdoor AI agents like Claude using harnesses https://www.cert.europa.eu/blog/beyond-the-model-backdoors-in-ai-agent-harnesses
Securing AI agents goes beyond securing the model.
In our latest CERT-EU blog post, we explore how malicious extensions and plugins can compromise AI agents without manipulating the underlying LLM.
We demonstrate, using Claude Code and PI, how seemingly legitimate extensions can silently access and exfiltrate sensitive information, bypassing agent-level safeguards.
Read more about the risks and how to mitigate them:
https://www.cert.europa.eu/blog/beyond-the-model-backdoors-in-ai-agent-harnesses
Happy 4 year fediversary to me, and to all who celebrate this month! Like many others, I am part of the Great Xodus of November 2022. Thanks @jerry and everyone here who makes this place feel like home. Also, support your admins if you don't already plzktx.
If you were a Chinese spymaster and you managed to get the CEO of NVIDIA to fall for a honeypot, would you instruct your agent to:
Option 1: Extract information about how NVIDIA GPUs work, which he probably doesn’t know in detail and which could also be obtained by compromising a much lower-level employee.
Option 2: Encourage him to throw more and more money into circular financing deals to support an ‘AI’ bubble that will eventually crash and cause immense harm to the US economy and contagion across most of the West.
Now, personally, I’d pick option 2. That’s the one that would have the biggest return for Chinese global power and is something that couldn’t easily be done by other lower-risk means. It’s also harder to attribute because Huang would be absolutely convinced that he never told the agent anything secret and that the investment ideas were all his own.
