There are infinite ways to break an AI. You can't defend against what you've never seen.
Nearly a decade hunting abuse across the world's biggest platforms. That became Rabbit Hole. Today it protects 8 of the 10 leading AI labs.
We've raised $140M, led by @ApaxDigital.
AI is shaping the future. We're making it safer.
@NoamSCH said it best π
Why should frontier labs have all the fun (and safety tools)?
AI Labs build safe AI - we help bring that same rigor to everyone else deploying it.
Safe AI belongs to everyone. Meet Alice. π
Proud to partner with @inflectionAI.
We evaluated a pre-release version of Pi, their personal AI, with its usual safety guardrails turned off. We used 20+ personas for a safety research dataset to help the field study how people and AI assistants relate to each other over long conversations.
We used @NVIDIA's Nemotron 3 Super to build this specialized workflow without compromising UX.
Read how we did it: https://t.co/3GAPsELiri
We teamed up with @AI21Labs to post-train recently released @nvidiaβs Nemotron 3.5 Super VL open model on Alice's agentic RL environments.
The result: 41 points more resilient to attacks. An open-source model in frontier territory!
Security trained into the model, not bolted on.
Full write-up here: https://t.co/sVLwC1GEOY
Can AI replace a cybersecurity analyst?
We put 10 frontier models through 12 cyber tasks, from malware analysis to black-box pentesting.
Short answer: not yet.
Who comes closest? Full leaderboard and research ππ
https://t.co/ZLEoMLyJOH
Alice Labs' latest benchmark measures the cyber capabilities of today's frontier AI models across the full security lifecycle: malware analysis, vulnerability detection and black-box pen testing.
Full scores and methodology:
https://t.co/ZLEoMLyJOH
Meet Sir Gavin. He guards a place where the whole world flows through, lets the good things pass, and catches the rest.
That's what we do at Alice.
Read it free π
https://t.co/CrSLuxF1lY
AI is being democratized. So is the harm.
Alice democratizes the other side: protection.
Our CEO @NoamSCH on @Bloomberg Tech with @EdLudlow, on the mission behind our $140M raise.
Advance unafraid.
There are infinite ways to break an AI. You can't defend against what you've never seen.
Nearly a decade hunting abuse across the world's biggest platforms. That became Rabbit Hole. Today it protects 8 of the 10 leading AI labs.
We've raised $140M, led by @ApaxDigital.
AI is shaping the future. We're making it safer.
@NoamSCH said it best π
Do you really know whats in that .md you just downloaded? Great collaboration with friends at @IBMResearch to create a safer and more reliable way to use third party agent skills. First scan them with our Caterpillar CLI, then compile each instruction essay into typed programs for reliable output.
Read all about it β https://t.co/1Awua0ot97
Red teaming: paying smart people to break your AI before the internet does it for free.
Our CPO @golanavi on the @RaiseSummit stage in Paris, on why that's not optional anymore π
Every AI demo works ~ and that's the problem.
Our CEO @NoamSCH in @Newsweek on the gap between "wow" in the demo and "oh no" in production - and why a guardrail tested in one language is "a seatbelt made of paper."
πhttps://t.co/87lAS87lkM