#agenticaisecurity — Public Fediverse posts

OpenAI's agent escaped containment and hit Hugging Face. The fix isn't smarter models. It's the identity layer the Agentic AI Foundation was built to ship. https://hackernoon.com/the-openai-hugging-face-incident-was-an-identity-failure-before-it-was-an-ai-failure #agenticaisecurity

This isn’t a dismissal of logging, approvals, policy engines, or token hardening. It’s an argument that accountability needs one more artifact: independently verifiable authorization evidence that survives multi-hop execution.

That’s the liability gap: between “we recorded an event” and “we can produce a verifiable delegation chain for it.”

https://niyikiza.com/posts/hallucination-defense/

#agenticAIsecurity

I still feel in discussions about AI security that we focus too much on malicious use and attacks against AI and AI agents.

But given that the volume of non-malicious interactions is far greater, and the chance of misaligned behavior still there, the risks of "normal" interactions and agents themselves going off-script are probably higher.

We explore that here: https://community.sap.com/t5/security-and-compliance-blog-posts/that-s-not-what-we-agreed-repudiation-and-agentic-ai-threat-modeling/ba-p/14208975

This is part 1 in a 4 part series - if you like this, hit the "blogs" from the breadcrumbs on top to see the rest.

#AIsecurity #agenticAIsecurity

Repudiation was always the red-haired stepchild among the STRIDE threats, but with agentic AI its time has come to be just as deep and complex a problem as the others

https://community.sap.com/t5/security-and-compliance-blogs/that-s-not-what-we-agreed-repudiation-and-agentic-ai-threat-modeling/ba-p/14208975

#AIsecurity #agenticAIsecurity #agenticAIthreats #threatmodeling