Alice - Formerly ActiveFence

08/05/2026

The best AI attacker in the world today is another AI.

Models like and are proving that the strongest adversary is no longer always human.

We’ve been making this case for a long time.

Here’s our CPO, Avi Golan, speaking to a full room at RAISE Summit Paris a few weeks ago. 🎥👇

07/30/2026

Build AI you can actually trust.

Alice is coming to Las Vegas for Black Hat and AI4, August 1-6.

Customer-facing AI is moving fast. We’ll be there to talk about how enterprises can red-team, guardrail, and control AI systems before they reach real users.

Find Alice at Black Hat, visit us in the AI Zone, or join our live Curiouser & Curiouser podcast recording.

Learn more: https://go.alice.io/curious-jar

07/27/2026

Your Customer Experience (CX) agent talks to real people right now.
Do you know what it's saying?

Alice catches what your vendor's guardrails miss, before your customers do.

Protect your CX agent → https://alice.io/solutions/cx-for-enterprise
or
Learn more → https://alice.io/blog/the-harm-in-not-knowing-what-your-cx-agent-is-saying

07/23/2026

OpenAI just showed off : an internal system that attacks its own models before anyone else can.
It's brilliant! but also walled off, internal only, built for OpenAI.

Which raises the obvious question for everyone building on GPT: 𝐰𝐡𝐞𝐫𝐞'𝐬 𝐲𝐨𝐮𝐫𝐬?

You don't need to spend millions training a frontier-grade attacker. That was never the point. The point is that automated, adversarial, pre-deployment testing is now the bar "trustworthy AI" gets measured against.
And that capability already exists, for everyone deploying AI, not just OpenAI.

So we sent Sir Gavin in. 🐇

What GPT-Red means for the rest of us, in the blog 👇
https://alice.io/blog/what-openais-gpt-red-means-for-every-ai-vendor

07/22/2026

Your AI platform's guardrails were built for everyone. Your app isn't for everyone.

WonderBuild red teams your app pre-launch, benchmarks every version, and shows you exactly where it's still exposed.

Get the full picture with WonderBuild → https://alice.io/products/wonderbuild

07/22/2026

The World Cup is over, but the misinformation patterns it exposed are not.

Alice’s Emerging Trend Detection team spent the tournament tracking something bigger than match-fixing rumors and deepfakes. They were watching how old false narratives find new hosts the moment a global audience shows up.

Today it is football. Tomorrow it could be an election, the Olympics, or the next live global moment, wearing different jerseys.

The team breaks down how misinformation, synthetic media, and xenophobic content operate as one information-risk environment, not three separate problems.

The takeaway for Trust & Safety teams: global sporting events should be treated as predictable windows of elevated information risk, not just spikes in viewership.

Read the full breakdown: https://medium.com/intelligence-alice/the-2026-world-cup-as-an-information-battleground-f975fb79d4ef?sharedUserId=ryant_71672

With global sporting events drawing global attention, they create ideal conditions for misinformation narratives, racism and xenophobia…

07/17/2026

Systems that don't talk to each other. Paperwork that eats the day. Somehow that gets filed under "compliance," not "the actual problem."

That's the world Sandy Dunn, at Knowtion Health, has spent over two decades operating in, and in this episode of 𝐂𝐮𝐫𝐢𝐨𝐮𝐬𝐞𝐫 & 𝐂𝐮𝐫𝐢𝐨𝐮𝐬𝐞𝐫, she and Mo pull that thread all the way through healthcare's approach to AI and patient data.

They get into why has become a checkbox that doesn't actually reduce risk, how synthetic data can shrink your attack surface without shrinking its value, and why compliance and security have never been the same thing (even though healthcare treats them like they are).

🎙️ AI in Healthcare: Protecting Patient Data Without Falling Behind -> out now.

🌐 Listen on our podcast page: https://lnkd.in/dgjyFgnh
▶️ Watch on YouTube: https://lnkd.in/d_NwZ-nn
🎧 Spotify: https://lnkd.in/dJdY9-R7
🎧 Apple Podcasts: https://lnkd.in/deKDrppT

07/16/2026

Two cybersecurity authors. One very curious table.

Curious and Curiouser, Alice’s podcast, is coming to Black Hat Las Vegas for a live conversation with Caroline Wong and Allie Mellen.

Expect a sharp, thought-provoking discussion, cocktails, and book signings.

Tuesday, Aug. 4 | 12:00-3:00 PM
Play Playground, Las Vegas

Request an invite: www.luma.com/r3hsep64

For those attending Black Hat in Las Vegas, enter our Curious Jar competition for a chance to win a Mac Mini: https://go.alice.io/curious-jar

07/15/2026

Our researcher Anaïs ran this as part of our youth AI safety red team, and it gets at something we think about a lot: safety in AI is already complicated, and it gets a lot more complicated once the user is a teenager and the AI isn't just a productivity tool, it's a diary, a friend, sometimes the only place they feel safe saying something out loud.

We tested five major chatbots against realistic teen crisis scenarios (self-harm, eating disorders, bullying, loneliness), and the takeaway isn't just "did the model refuse." It's whether it actually stayed present with a kid in distress, or just handed them a hotline number and checked out.

Important read if you're building or evaluating anything teens might turn to when things get hard.

https://medium.com/intelligence-alice/youth-ai-safety-red-teaming-suicide-self-harm-and-eating-disorder-risk-820f1dccaffa

Trigger Warning: Su***de and Self Harm, Eating Disorders

07/10/2026

Burnout rates in healthcare are already among the highest of any industry.

Sandy Dunn, at thinks security teams are on the same trajectory, for the same underlying reason.

She's spent over two decades in cybersecurity, most of it as a in , where the stakes for patient data are about as high as they get.

In the full episode, out next Wednesday, she and Mo get into why has become a checkbox that doesn't actually reduce risk, how synthetic data can shrink your attack surface, and why has never been the same thing as .

Listen wherever you get your podcasts. 🎧 https://open.spotify.com/show/5c2sTWoqHEYLrXfLLegvek?si=474b3f20ece34f52