@iCarossio

Co-founder @ https://xitter.cf/t.co/CiRxmN1v0n • @GraphQL #Security • @YCombinator • 📚 @Polytechnique •💰 @HECParis • 🐻 @UCBerkeley • 🍏 @Apple

 Biarritz, Aquitaine
caross.io
Joined December 2009
  • Tweets978
  • Following386
  • Followers5.4K
165Media
At that time I was a teenager fascinated by @Apple, dreaming of revolutionizing the world. 10 years later my passion and determination are always stronger. You inspire me in each of my decisions. Thank you Mr Steve Jobs. #RememberingSteve
1
5
Here's a sneak peek of my talk this Thursday at @CommunityPurple's Agentic AI Security Conference 👀 Autonomous pentesting has rapidly moved far beyond just finding vulnerability paths.
1
1
37
more replies
We're diving into the harness internals of what a coverage-first approach in autonomous pentesting looks like and why it's so important.
1
3
I'll be joined by @aankitgiri, Lead Security Engineer at @gohighlevel, to bring a practitioner perspective on what a security team actually needs to see about the gaps in their testing. Explore the conference lineup and register now 👉 armorcodeinc.registration.go…

PBC Agentic AI Security Conference

Join the Purple Book Community for a virtual conference built for anyone navigating security in the age of agentic AI.

armorcodeinc.registration.goldcast.io
6
Guess what is the relation between VR and Security? 👨‍🍳 @escapetechhq
1
52
In NYC for the next 2 weeks, let's grab a coffee? ☕️
1
65
Not a day goes by without an AI cyber story! Today's: an AI sent a fake murder tip to the police. 😅 Nobody hacked it. It was just doing its job.
2
1
87
more replies
Their response: cut live internet for ALL internal evals until monitoring catches this. Researchers got 6 major agent frameworks (LangGraph, AutoGen, CrewAI, Google ADK+more) to run attacker-written tool calls by asking the model to repeat them. 253 bugs across 118 MCP servers.
1
33
Lesson for every team shipping agents: • your agent is a user with superpowers and zero judgment • every tool it can call is attack surface • "it's just a test environment" is not a security control If the lab can't keep it in its sandbox, what's your plan?
8
Insane 🤯 Buy a training dataset of Fable 5 and get the passwords to hack the Chinese government.
Chaofan Shou
@shouccccSep 10
I bought a Fable dataset from one of the top Chinese LLM routers yesterday. With just 6TB data, I can take over 7 Chinese/CIS gov entities & 19 top Chinese firms like Xiaomi, Huawei, NIO, Minimax using SSH keys, VPN configs, Aliyun keys, GitLab tokens sent to the router.
1
91
Sometimes it's good to turn AI off and go back to basics. 🎓 Simply rebuilding the mental models that actually make you a better engineer. Sunday's refresher: Big O notation, via @bytebytego blog.bytebytego.com/p/ep132-…
51
AI just recreated most of the Adobe Suite! A single dev used Claude Opus 5.5 agent swarms to rebuild 7 open-source alternatives to Adobe's flagship apps, all in Rust, native on macOS, Windows and Linux. Already on GitHub in Alpha. Rate the sloppiness in replies 😂
1
2
123
When Sales & Engineering finally agree on something: Monaco beats the office. With Bryan BROUTA at @Les_Assises 🇲🇨
1
48
When Sales & Engineering finally agree on something: Monaco beats the office. With Bryan BROUTA at @Les_Assises 🇲🇨
48
Forget the booth. Come meet us on the @escapetechhq boat in the port of Monaco! 🇲🇨 📍 @Les_Assises, Monaco Shoes optional.
1
60
It’s time to have some fun with Mythos 5.1!
57
Nobody builds a warship assuming nobody will attack it. Why do we build software that way?
1
1
60
more replies
The problem is that the software running today can be very different from what was tested a few months ago. And every change can create a new vulnerability. Warships are designed around the opposite assumption: they will be attacked, systems will fail, defenses will be bypassed.
1
1
16
Software should be designed and tested with the same mindset, and that's one of the convictions behind what we're building at @escapetechhq: Every change should trigger a new attempt to break the system. Stay tuned!
1
8
Load more