⚡🛡️ Evan Pappas

🛡️ Ex Technologia Libertas - Έλευθερία διὰ τῆς τέχνης - (Dec/Acc)

[⚡🛡️ Evan Pappas @Hevalon]

@mfpiccolo roger. I'm connected there now.

[⚡🛡️ Evan Pappas @Hevalon]

@mfpiccolo thanks, connected. i don't think the verification is working though.

[⚡🛡️ Evan Pappas @Hevalon]

$ claude "list all the vulnerabilities in my cluster"

[Ado @adocomplete]

Introducing Claude Fable 5, our most capable public model ever.

Best-in-class for software engineering, scientific research, knowledge work, and vision.

Available today on all paid plans, in Claude Code, on the Claude API, and all major cloud platforms.

[⚡🛡️ Evan Pappas @Hevalon]

@mfpiccolo added on my todos for this week :)

[⚡🛡️ Evan Pappas @Hevalon]

@mfpiccolo I really like the Control policy semantics you guys have made. I'll spend some time to experience your setup.

One more thing to have in mind regarding access control is the goal intent.

I'm writing more about it here https://t.co/4FW1nHpYTW

[⚡🛡️ Evan Pappas @Hevalon]

@mfpiccolo nice one, how are the control policies managed?

[⚡🛡️ Evan Pappas @Hevalon]

@addyosmani with great power, comes great.. vulnerabilities. IMO Secure control policies need to be part of the control plane of the loop orchestrator https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@ritu_twts coding is power

[⚡🛡️ Evan Pappas @Hevalon]

@AiCamila_ spot on!

[⚡🛡️ Evan Pappas @Hevalon]

@shannholmberg nice one, I think the control security aspect of agentic loops is not covered as much and it's quite important. I'm trying to cover more about it https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@Voxyz_ai nice one, one additional major point is about the control security aspect of agentic loops, I'm expanding more about it here https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@cellinlab nice one, i'm also expanding on the control security aspect of agentic loops here https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@Krishnasagrawal one layer that's missing out, and it's quite pivotal IMO, is the control security around the agentic loops. I'm expanding more about it here https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@_avichawla one pivotal point, is how a control security system around the agentic loops can be implemented https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@PawelHuryn one missing point (major IMO) is how you build a control security around them, i'm expanding more about it here
https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

@akshay_pachaar nice one, i'm expanding on the control security aspect of agentic loops here https://t.co/4GbkEvkaYC

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

The rule is simple: the model can propose an action, but something above it has to decide whether it runs.

Writeup and code: https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

How can you make sure your agents don't perform a disruptive action, when they run on a loop, like deleting your production database?

I built a system to stop that; Security Control for Agentic Systems

https://t.co/7b5W6gbIzK

[⚡🛡️ Evan Pappas @Hevalon]

One check has no research behind it: refusing to route a request to a model outside the region its data is allowed in. The routing papers all optimize for cost and quality; none of them ask where the data is allowed to go.