Mud

[Thomas Roccia 🤘 @fr0gger_]

🤓 Sneak peek from our training session in Sydney last week!

A deep dive into AI Threat Intelligence, agent security, adversarial prompts and the AI threat landscape.

The next session will take place at @BlackHatEvents USA this August.

See you there!

👉 https://t.co/bnNPBSxgAP

[Thomas Roccia 🤘 @fr0gger_]

🤓 Soon enough (if not already), you will have to investigate AI breaches and answer these questions:

How do you hunt for adversarial prompts?
How do you investigate a breach in your AI agent's execution?
How do you detect that your agent has been compromised?

I have been working on these topics for a while and I have already investigated multiple agent compromises.

Now it is time to make this into a formal security practice!

[Thomas Roccia 🤘 @fr0gger_]

🤩 We just wrapped up 2 days of my training Practical AI for Threat Intel in Sydney!

It was packed and the class was fantastic!

If you want to step up your skills and learn faster from our latest research, the next session will be hosted at @BlackHatEvents in August!

https://t.co/uq7xEUfWlg

[Thomas Roccia 🤘 @fr0gger_]

We reported this prompt two days ago in PromptIntel! Have a look 👇

https://t.co/F6c08dXeeQ

[Thomas Roccia 🤘 @fr0gger_]

With the @SLEUTHCON swag I really feel
I am part of a gang 😈

[Thomas Roccia 🤘 @fr0gger_]

Also straight from the Anthropic documentation 👇

[Thomas Roccia 🤘 @fr0gger_]

Let me explain what Boris means by "creating loops instead of prompts" because I think it confuses a lot of people!

A loop is a task or an objective given to an agentic workflow or agent. But instead of having multiple back and forth interactions with an agent, you define a goal and let the agent iterates through a series of steps, decisions and prompts until the task is complete!

To make it clear your agent still works with text, code, files and other inputs and sorry but your loop/goal still needs a prompt.

The difference is that instead of sending a prompt and getting a single answer back, the agent uses that prompt as an objective and executes multiple steps until the task is complete.

However what is also possible is to trigger a loop based on an action rather than a chat interaction.

For example, you drop a file into a folder and your agents automatically start the workflow you defined for that file: analyze it, extract information, enrich the findings, generate a report, and so on.

Hope this makes it a bit clearer 🤓

[Thomas Roccia 🤘 @fr0gger_]

New coin unlocked for my collection! Thank you @SLEUTHCON this one will have a special place 🐍

[Thomas Roccia 🤘 @fr0gger_]

🤓 This morning at @SLEUTHCON, I talked about how AI is being targeted and leveraged by cybercriminals.

Which is beyond simply using models in their operations, attackers are also actively targeting AI environments themselves.

That AI agent you trusted inside your organization is becoming a prime target because you don’t know what it is doing while it is running.

And attackers know it. 💀

[Thomas Roccia 🤘 @fr0gger_]

🤓 On Friday I will have the honor to present the keynote at @SLEUTHCON! Come say hi if you are around!

[Thomas Roccia 🤘 @fr0gger_]

The best recent feature of VScode 😊

[Thomas Roccia 🤘 @fr0gger_]

🤓 Adversarial Prompt (IoPC) available in PromptIntel including Nova rule for detection!

https://t.co/MicKXwk71d

[Thomas Roccia 🤘 @fr0gger_]

Interesting behavior 🤔

[Thomas Roccia 🤘 @fr0gger_]

🤓 PromptIntel the database of Adversarial Prompts or Indicator of Prompt Compromise, gets a redesign!

Check this out 👇

https://t.co/4ko4p0A3v9

[Thomas Roccia 🤘 @fr0gger_]

Even if I code less than before I still enjoy crafting Python scripts by hands. Just because I love this language 🤓

[Thomas Roccia 🤘 @fr0gger_]

@cuctrosi You can check PromptIntel here: https://t.co/4ko4p0A3v9

[Thomas Roccia 🤘 @fr0gger_]

https://t.co/gytntwWrob

[Thomas Roccia 🤘 @fr0gger_]

🤓 @cuctrosi just releases a bunch of Nova rules to detect Adversarial Prompts (IoPC) referenced in PromptIntel!

Nova is an open source Prompt Pattern Matching, you can use it for prompt hunting and AI guardrails!

Check the latest rules created👇

[PentesterLab @PentesterLab]

If you are in France 🇫🇷 and want some stickers:

https://t.co/AYPrYXemik

[Thomas Roccia 🤘 @fr0gger_]

Still couple days to vote 👇