Sunday Edition #13: The Hidden Layer of AI Success
Security as an accelerator, AIUC-1 status, and Fin Meetup expansion
A big week for AI’s hidden layer: safety. Security used to be a bottleneck. But in the AI age, safety is what accelerates productivity.
You can’t run a bullet train on rusted tracks. If you want to move customers at the speed of AI, you need infrastructure that can handle velocity without derailing.
A new standard for trust
We started the week with a major milestone: Fin achieved AIUC-1 certification.
You might be thinking, “AIUC-1? Those are letters & numbers alright, but what do they mean?”
AIUC-1 is the first security standard designed specifically for AI agents, essentially acting as a SOC 2 for AI.
To get it, Fin had to undergo independent auditing and quarterly “adversarial testing” across thousands of risk scenarios. As Chief Information Security Officer, Thibault Candebat noted in the announcement:
“As AI becomes more deeply embedded in customer-facing work, it’s essential that businesses can rely on systems that are safe, reliable, and governed to the highest standards.”
For our customers, this provides independent validation that our systems are safe, resilient, and enterprise-ready. Darragh shared a bit more on LinkedIn.
This is one of many steps toward redefining “safe” for the next generation of agents. Read the full announcement here.
From “security” to “trust”
We also launched a new podcast series to highlight conversations with leaders in the space. Subscribe on Apple Podcasts and Spotify.
In last week’s Fin Meetup episode, Thibault joined Vanta’s Head of Engineering EMEA, Todd Royal to discuss how customer interactions are evolving.
According to Thibault, there’s a shift:
“We’re moving from security questions to more trust and safety questions. You have this new class of risk with AI hallucination. Now you have to build controls for that kind of thing as well.”
Todd pointed out that this isn’t all about avoiding risk, it’s about business advantage:
“Compliance is no longer about checking the box. Continuous monitoring and having insight into ways that AI might be able to accelerate your security posture, accelerate your business. Those are things that customers want as well.”
And if you’re more of a viewer than a listener, you can watch the full discussion on our YouTube channel.
Entering the simulation
The practical side of this “trust” conversation is simulations.
If you’re going to put an AI in front of your customers, you need to know what it’s going to say before it says it. Thibault highlighted how customers are now running simulations to test Fin’s responses against predefined inputs.
“This helps customers validate that the agent is working according to their requirements before deploying it to production. It also allows them to identify if anything unusual is happening on Intercom’s side.”
It’s the difference between hoping for the best and proving it.
This week on Ideas
More exciting updates from the week.
💰 Six figure startup support. We released an AI Startup Packet worth over $100,000 of AI credits.
🌎 Blueprint world tour. Amsterdam, Paris, Berlin, and a sold-out London finale; we held CX talks, product demos, and Fin Voice live-sessions to help our customers scale.
🏺Clay doesn’t fit the marketing mold. Paul Adams met with Clay Co-founder, Varun Anand to discuss non-traditional ways they go to market; view the recap here or watch the full conversation below.
Our parting thought
In the old world, security was a gate. Today, security is the engine. You can only automate what you can trust.
See you next Sunday.
•••
Great reframing of security from gate to engine. The simulation point is what really matters though becuase it shifts validation from post-deployment firefighting to pre-deployment confidence building. I've seen teams struggle with customer-facing AI where the simulation layer was an afterthought and they ended up with response drift nobody caught until production. Running those adversarial testcases upfront is the difference between controlled deployment and hoping for teh best.