Days to Seconds: Harnessing Confidential AI Agents

Welcome to AI Confidential, your biweekly breakdown of the most interesting developments in confidential AI.

Today we’re exploring:

• Why trust is the cornerstone of enterprise software

• The concerning rise of “vibe hackers”

• Open source projects we’ve been following

Also mentioned in this issue: Kellie Romack, Clayton Boehm, Jason Lazarski, Soriya Ngo, Steve Miranda, ServiceNow, IBM, Air AI, Anthropic, GoLaxy, Gartner, The Economic Times, McKinsey & Company, SF Tech Week, Oracle, Google DeepMind, Skyfire, Tencent, Sentient, and ByteDance.

Let’s dive in!

Trust is the heart of enterprise software—especially in the age of agentic AI.

In this week’s episode of The AI Confidential Podcast, we sat down with Kellie Romack, Chief Digital Information Officer at ServiceNow, to unpack a rising issue for enterprise leaders:

Can we deploy AI agents without sacrificing privacy, security, or control?

If you ask Kellie, the answer is simple:

Yes, if they’re confidential from the start.

Instead of treating AI like a black box, confidential agents create a glass box. Data stays encrypted—even while it’s being processed—and every interaction produces an auditable trail that proves policies are being enforced.

The result? Trust isn’t assumed. It’s verifiable.

When you put this into practice, the benefits start to cascade.

At ServiceNow, Kellie and her team partnered with OPAQUE, Microsoft Azure, and NVIDIA to deploy confidential AI agents—and now their sales commission workflow that once took four days takes just eight seconds. 

With one simple change, thousands of employees now get instant answers and support staff gained back valuable time to focus on more strategic work.

Kellie explains it best:

“We’re obsessed with building and keeping trust, and confidential AI helps us balance personalization [and] privacy — so we can deliver . . . those better experiences . . . completely integrated on our platform end to end.  We're embedding confidential AI agents into HR, into finance, into legal, into our go-to market.”

ServiceNow is laying down a reference architecture other enterprises will follow as they scale their own AI strategies.

This foundation of trust is what makes agentic workforce management possible.

AI agents partnering with humans across departments to amplify talent, boost productivity, and unlock collaboration at scale—and we’re still only in the early days.

Also in this episode, we chatted with Kellie about:

• Why secure agents are now the baseline for enterprise AI

• How the Zero Service Desk boosted team capacity by 40%

• Why breaking down silos is key to unlocking enterprise-wide value

Visiting the ServiceNow headquarters in Santa Clara was an absolute delight. 

Listen to the full episode with Kellie now.

Keeping it Confidential

What percent of companies use genAI tools?

1. 35%

2. 50%

3. 65%

4. 80%

See the answer at the bottom.

Code for Thought

Important AI news in <2 minutes

🔐 97% of enterprises hit by AI-related cyberattacks lacked proper AI access controls, with many having zero governance policies, an IBM report revealed. 

⚖️ The Federal Trade Commission (FTC) is suing Air AI for deceptive marketing, citing claims the startup’s AI agents could replace employees. 

🔮 Using Anthropic’s Clause, “vibe hackers” are extorting content, orchestrating romance scams, and executing psychologically-charged attacks across all major industries.

📰 GoLaxy, a startup with ties to the Chinese government, found to have used AI to generate propaganda campaigns in Hong Kong and Taiwan, according to new Vanderbilt University research.

💻 AI is fueling an uptick in cyberattacks across India, including deep fake scams that cost mere pennies to create, the Economic Times reported.

Community Roundup

Updates involving OPAQUE and our partners

🎉 Exciting news: a free webinar event is coming next week!

Join members of the OPAQUE team for an educational 30-minute session, all about confidential AI for insurance.

When:

⏰ Wednesday, September 17th, 1pm ET | 10am PT

About what:

✏️ How OPAQUE protects sensitive insurance data in AI workflows

✏️ Real-world use cases for confidential AI adoption

✏️ Plus, a practical demo showing OPAQUE in action!

RSVP now to get added to the list.

OPAQUE events

SF Tech Week is happening October 6-12, and honestly? We couldn't be more excited. 

This year's lineup is fully stacked—with founders, VCs, and tech leaders from around the world gathering in the heart of Silicon Valley for a week of next-level networking.

And, if all of that wasn’t enough, our own team is hosting not one, but two events on Tuesday, October 7th:

1️⃣ Agentic AI for the Enterprise: What's the Catch?
 11:30 AM - 1:30 PM

Everyone's talking about AI agents, but let's cut through the hype. For this panel, we've got speakers from Google DeepMind and Skyfire sharing what's actually working in the real world (and what isn't). 

2️⃣ AI Confidential Dinner
6:00 - 8:00 PM

We’re throwing a dinner for founders, builders, and operators who ship AI products. We promise no presentations and no pitches—just good food and even better conversations.

💌 Ready to join us? Register now to secure your spot.

OPAQUE in the wild

Exciting news—OPAQUE was mentioned in the latest Gartner report! 🎉 

In their new report, Emerging Tech: Confidential AI Drives Innovation and Unlocks New Opportunities, OPAQUE is mentioned in multiple key use cases, including:

• How OPAQUE acts as a privacy layer on LLMs, ensuring sensitive data never reaches any model

• How Accenture’s investment in our platform highlights the emerging importance of confidential AI across all industries

Check out the report—it’s filled with tons of useful insights about securing genAI workflows and protecting sensitive data used by AI models, so it’s worth a read!

Open source spotlight 

🌐 Chinese tech titan Tencent unveiled Hunyuan-MT-7B and Hunyuan-MT-Chimera, a joint model that translates text across 33 languages, outperforming big hitters like OpenAI and Google. 

🤖 AI startup Sentient launched The GRID, an open-sourced agentic marketplace consisting of more than 40 agents that can coordinate calendars, plan travel, and visualize data.   

🖼️ ByteDance released USO, an image-editing model that uses text-based prompts to generate images with high subject and identity consistency.

Quotable

💼 “Agents will redefine the boundaries of execution, reshape the role of human capital, and introduce a new layer of embedded intelligence that will become central to business value creation.”

 — Steve Miranda, Oracle’s Executive Vice President of Applications Development

Trivia answer: 80%

According to a report from McKinsey & Company, nearly 80% of companies use genAI tools. Interestingly, many report no positive impact on their bottom line, with many projects getting trapped in the pilot stage without proper data security and confidentiality in place.

Stay confidential!

- Your friends at OPAQUE

ICYMI: Links to past issues

• What AI’s Most Innovative Thinkers are Saying

• Agents are the New API Client

• 3 New Ways to Tackle AI Security