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

Today we’re exploring:

• Predictions for the enterprise AI market

• The explosive popularity of AI-generated personalized dolls

• New open source projects you need to know about

Also mentioned in this issue: Sriram Raghavan, Mark Russinovich, Sam Altman, Chris Gibson, IBM, BeeAI, Docling, OpenAI, SS&C Technologies, Solutions Review, Snowflake, AllThingsOpenAI, Google, Microsoft, NVIDIA, Intel, Ember, Alibaba Cloud, and Deep Cognito.

Let’s dive in!

Season 2 of The AI Confidential Podcast continues this week—chock-full of bold predictions about the future of enterprise AI.

A few weeks ago, hosts Aaron Fulkerson and Mark Hinkle sat down with Sriram Raghavan, the Vice President of IBM Research AI, at the AllThingsOpen.ai conference in Durham, NC. 

As an experienced industry expert, Sriram leads a global team of over 750 research scientists and engineers focused on advancing AI.

Innovation is always at the forefront of his work—which is why IBM’s recent successful launches like BeeAI and Docling were created using two non-conventional approaches:

1. All IBM AI projects are built with the mandate of open innovation

2. IBM is leaning into smaller “fit-for-purpose” models over only building LLMs

This shift in approach is worth paying attention to, because it challenges the current market.

Better yet, these approaches open the door for specialized innovation across things like model architecture, blending, and data modalities…

All of which Sriram predicts are major untapped axes of innovation.

(These are just a few of the juicy tidbits we get into during this episode. The man is truly a well of industry wisdom).

In the episode, Sriram also digs into:

• What innovations he sees coming for enterprise AI

• Insights into the latest launches from IBM

• How his team is approaching responsible, secure AI

You won’t want to skip this one. You can listen and subscribe here.

Code for Thought

Important AI news in <2 minutes

🧸 People are turning themselves into lifelike dolls and action figures, reigniting discussions of the value of human-made art.

🤖 78% of enterprises using AI agents don’t fully trust the technology, according to a new SS&C report.

🛡️AI agents will soon carry out the next iteration of cyberattacks, experts warn.

⚙️ OpenAI CEO Sam Altman trolls his own company’s naming conventions as they launch GPT 4.1, the faster, better successor of GPT-4o.

💸 Companies are seeing a 41% ROI on generative AI initiatives, new Snowflake research finds.

💡71% of early adopters likely can’t afford every generative AI use case they’re eager to explore, the same report claims.

Community Roundup

Updates involving OPAQUE and our partners

🚨 Attention AI, data, and security leaders

There’s still time to register for The Confidential Computing Summit—the premier conference on confidential AI. 

• Taking place in San Francisco on June 17–18, the two-day event highlights cutting-edge advancements in securing AI workflows.

• You’ll meet the builders, buyers, and innovators shaping AI infrastructure, including speakers from industry giants like Google, Microsoft, Intel, and NVIDIA.

Plus, as an added bonus, join us June 16th for an exclusive workshop, NORMALizing AI: Deploying Enterprise-Grade Intelligence, presented in partnership with All Things AI.

You don’t want to miss it—register here to get your ticket.

OPAQUE in the press

Yesterday, we were excited to announce that our secure AI solution is now available on the Microsoft Azure marketplace.

Here’s what Mark Russinovich, CTO, Deputy CISO, and Technical Fellow, Microsoft Azure, had to say about the partnership:

“By supporting innovative solutions like OPAQUE's Confidential AI Platform through our Pegasus Partner Program and the Azure Marketplace, we're empowering enterprises to accelerate their AI initiatives while upholding stringent security and compliance standards."

Read the full press release here to learn more.

Open source spotlight 

🧠 Ember is an LLM framework that coordinates models with different reasoning speeds to improve accuracy and reduce cognitive overload, preventing “overthinking.”

🤖 Alibaba Cloud’s Qwen2.5-Omni-7B aims to power cost-effective AI agents with a multimodal LLM that processes text, audio, images, and video on edge devices.

⚙️ Deep Cogito’s Cogito v1 is an LLM with hybrid reasoning, designed to self-improve without human input, marking a push toward superintelligence.

📢 Got an open source project you’d like to share? Respond to this email detailing your work, and we’ll consider including it in the next newsletter. We’d love to hear from you!

Quotable

🔐 “Success in 2025 will come to those who balance AI’s potential with the vigilance to address its risks, all while keeping data privacy at the forefront.”

— Chris Gibson, CEO and Executive Director at the Forum of Incident Response and Security Team. (Source: Solutions Review)

Until next time, keep it confidential!

- Your friends at OPAQUE

ICYMI: Links to past issues

• Unleashing the Power of Agents

• Why Data Masking Isn’t Working for Your AI

• The AI Bottleneck is Trusted Data