6 results for “jkirk”
LLMs can contain nasty backdoors AND YOU CAN'T FIND THEM
Any open model whose weights have been edited can carry a backdoor, whether it's a task-specific fine-tune, a merged adapter or an abliterated build. We built one to see what it takes,
https://projectdiscovery.io/research/how-abliterated-models-can-get-you-pwned HT @jkirk
#AI
Amidst the personal AI agent boom, Citibank and Wells Fargo are advising customer to not enter credentials, MFA codes, financial data or documents into AI tools. Citi: "AI assistants can make mistakes, and that means you could be liable for errors it makes."
Abliterated open-weight models could have backdoors. Possible to detect by diffing against the base model or implementing runtime controls: https://projectdiscovery.io/research/how-abliterated-models-can-get-you-pwned
Re: the ConnectWise incident. App Instance Lock is the default in Entra ID as of June 2026. It's a good idea to go back now make sure it is enabled for multi-tenant apps https://learn.microsoft.com/en-us/entra/identity-platform/howto-configure-app-instance-property-locks
Also, this his video from really captures why the reach of the ConnectWise incident is wide - one single secret replayed against heaps of tenants running this highly privileged app. https://www.youtube.com/watch?v=Q36aL_HIS2w
It is such a deluge of hard news these days, and I find myself not wanting to add to it. But I want to close the loop on this one.
Many of you shared my post about my friend’s contractor, who was kidnapped by ICE last Wednesday on his way to his immigration appointment.
Taken Wednesday. Sent to Louisiana Thursday. On Friday, he called his family from Honduras, the country he left as a small child.
It’s over. No appeals.
