GOVERNANCE

The case against paying for AI with your data

If the AI industry has a counter-cultural movement right now, then it's the group focused on decentralizing power and enabling bottom-up innovation. 

On Tuesday at the Confidential Computing Summit in San Francisco, a group leaders from major tech companies, including Google, Microsoft, AMD, Intel and others, came together to spearhead efforts to implement data sovereignty, cybersecurity, and privacy, putting each enterprise in control of its own destiny so it can thrive when the unstoppable wave of AI agents hits in the months and years ahead. 

A lot of the answers on how to make that happen circled back to open-source solutions, which wasn't surprising, considering The Linux Foundation was the organizing entity. 

"In order to use the best models, people are giving up their data, and that data then pulls into one or two [models]... And then everybody becomes dependent upon one or two entities," said Aaron Fulkerson, CEO of Opaque and emcee of the event. "That creates a system that is very fragile and brittle. It's one in which there's a single off-switch, one in which there's a single set of blind spots, and it is not a durable environment."

The diverse set of speakers made recommendations that coalesced around four key themes:

• Protect what agents hold in memory: Organizations need mechanisms to encrypt, control, and even permanently delete what agents learn and remember, especially when handling confidential customer, financial, healthcare, or other strategic data. 

• Build systems with hardware-based trust: Enterprises can't trust software policies to govern increasingly autonomous agents. The speakers argued that trust must be rooted in hardware-based confidential computing, where security rules are enforced and verified cryptographically at the chip level.

• Give every agent a verified identity and audit trail: Every agent operating in an enterprise environment needs a tamper-proof record of what it's allowed to do, what it actually did, and what data it touched. This makes agent behavior auditable, accountable, and defensible to regulators. At the event, The Linux Foundation announced the Agent Name Service (ANS), much like the DNS system that provides a similar function for the internet. 

• Control data sovereignty with external models: Multiple speakers warned against a future in which enterprises must hand over their most valuable data to a handful of AI providers to access the best models. The alternative they advocated is a "sovereign AI" approach. That means holding your own encryption keys, setting your own policies on how data flows, and getting verifiable proof that those policies were followed. Without this, companies risk becoming permanently dependent on a handful of AI providers.

"We should never have to pay for AI with our data," said Fulkerson. "Let's optimize for freedom: the freedom to choose which model you want without having to pay for it with your data."

The group of companies that came together at the Confidential Computing Summit articulated clear, comprehensive insights and recommendations for creating decentralized, locally controlled AI. The pendulum swings between centralization and decentralization doesn't just mirror the on-prem vs. cloud debates a generation ago — although it would be easy to see it that way. AI is a much more active agent (forgive the pun) in the future of technology than cloud storage and hosting ever were. The stakes are far higher for both enterprises and society. So it's encouraging to see the intensity and the precision of the recommendations coming out of this event. Notably absent were OpenAI and Anthropic, although Anthropic's deputy CISO is on the panel I'm on Wednesday at the event. Follow my updates in real-time on X/Twitter at x.com/jasonhiner to see what more we learn. 

TOGETHER WITH NICE

Wrong order. Right resolution. See how CX AI delivers for Kristen Bell.

See how NiCE CX AI helped make Kristen Bell's customer service problem disappear.

NiCE CX AI helps leading brands proactively resolve issues, personalize service, and deliver seamless experiences at enterprise scale.

By connecting intelligence across every interaction, brands can anticipate customer needs, accelerate resolutions, and continuously improve customer experiences.

Watch how

PRODUCT

Why Claude wants a new role: colleague, not chatbot

Anthropic wants Claude to be a teammate, rather than a tool. 

On Tuesday, the company unveiled Claude Tag, a system that embeds its flagship model into Slack, allowing it to "join as a team member." This allows users to grant the model access to certain channels, tools or codebases, and delegate tasks to it by tagging Claude. 

Anthropic noted that Claude builds context by remembering relevant information from the channels it's a part of, and uses it to plan and complete future tasks. 

Claude Tag is currently available on Slack for Enterprise and Team customers and will expand to other work platforms in the future. In its blog post, Anthropic pointed out a few advantages of this system: 

• The tool is "multiplayer," meaning that it interacts with everyone in the channels it's in, and anyone can transparently see what the model is working on, too. It can also pick up conversations where others left off. 

• Claude Tag learns over time by following along with the channels it's in and building context about an organization, and it automatically learns from other channels and data sources when given permission. This negates the need for users to "explain things to it from scratch over and over again," Anthropic said. 

• If "ambient" behavior is enabled, it keeps users apprised of information it thinks they need to know, and flags relevant information from different channels and tools. 

• And it works asynchronously, allowing users to give Claude a task and allow it to work in the background. The tool can also schedule tasks for itself.

The company said that it sees Claude Tag as an evolution of Claude Code, and that it's "one of the main ways we get things done at Anthropic." In its blog post, the company said that 65% of its code is created by an internal version of the product. 

"The same pattern is now spreading well beyond engineering—we’re tagging Claude to chase down product metrics and data, work through support tickets, or even help find the root cause of tricky bugs," Anthropic said in its blog post. 

It's not the first time that Anthropic has sought to extend the success it's seen with Claude Code to other domains. In January, the company launched Claude Cowork, which gives knowledge workers an agent to autonomously handle tasks when given access to specific tools or folders. More recently, the company unveiled Claude Design, an AI tool that allows designers to iteratively build visual products using the chatbot.

Anthropic leaning into Claude Tag is about more than just roping in more enterprise users. It marks an evolution in how enterprises and organizations view this technology. Anthropic positions Claude Tag as a digital teammate, akin to human coworkers, rather than a tool. Additionally, the fact that this is built into Slack, one of the most popular platforms for enterprise communication, further embeds that concept within businesses. However, this mindset shift brings up two potential concerns. For one, the more we think of these tools as coworkers, not as machines with built-in flaws in accuracy and data security, the more we must trust them, and the more risk we undertake at the same time. Additionally, this shift risks further growing the narrative that many experts warn against: that AI is ready to automate jobs and replace employees, rather than augment and empower them.

TOGETHER WITH KALSHI

The First Regulated Perps Market in the US Is Live

Perpetual futures (perps) are now live on Kalshi.

Offshore, perps do ~$90 trillion in volume annually — until now, none of it was legal in the US.

Trade on whether an asset's price goes up or down. No expirations, no rolling costs, no complex payoffs.

Try Kalshi Perps today.

WORKFORCE

AI raises the bar for human judgment at work

It’s a tough time to be looking for a job in the age of AI. But executives say workers who can adapt to the times still have an edge.

During a panel at VivaTech in Paris, business leaders discussed how AI is reshaping hiring. Sue Duke, LinkedIn’s head of global policy and economic graph, pointed to internal data showing that hiring across Europe is down 25% from pandemic-era highs and 15% year over year. In the U.S., LinkedIn data shows hiring remains 24% below pre-pandemic levels and was down 6% year over year as of March.

Even as hiring slows, employers say the roles that remain open increasingly demand a new mix of hard, soft, and AI skills. Companies are prioritizing candidates who can use AI tools effectively while bringing strengths that are harder to automate. For many employers, that means knowing when to use AI, how to evaluate its outputs, and how to integrate it into existing workflows.

"We see the most resilience in hiring in roles that combine AI efficiency and unique human skills," Duke said during the panel. 

Because fixed skill sets age quickly, human skills are becoming more important. Duke said adaptability and curiosity are essential to getting hired as AI reshapes the workplace. LinkedIn estimates that 70% of the skills used in most jobs will change by 2030, with AI acting as a major catalyst. As job requirements evolve, workers who can continuously learn and apply new technologies have an advantage.

Software engineering offers a glimpse of what that shift looks like in practice. Before generative AI, employers largely hired engineers for their coding ability. Today, tools like Codex and Claude Code can generate high-quality code, pushing companies to look beyond technical proficiency.

Employers want engineers who can work alongside AI and help organizations integrate it effectively, said Ruth Harper, SVP and chief marketing and sustainability officer at Manpower Group, a global talent recruitment agency. That requires cross-functional collaboration, change management, and the ability to bring colleagues along.

"How do you drop AI into the workflow? How do you help your organization get comfortable with agents working next to your humans, for your humans? How are they part of the team? That is not a technology job," Harper said.

Still, leaders emphasize that hard skills like coding remain essential.

"The combination [of skills] is what can be absolutely critical," Harper said. "The technical skills are a given, but if you only have technical skills, somebody else is going to get to the top of the list first."

The bar for landing a job keeps rising. As AI reshapes work across industries and experience levels, employers are looking for candidates who can adapt alongside it. Specialized expertise alone is no longer enough. Workers are increasingly expected to pair technical skills with emotional intelligence, strong communication, and a willingness to learn. In a world where more tasks are being automated, the most valuable employees may be those who can demonstrate they have deeply human advantages.

LINKS

• Oracle has cut 21,000 in the past year, almost 13% of its workforce

• Meta unveils new line of AI glasses, powered by Muse Spark model

• Qualcomm nears deal to acquire AI software firm Modular

• Google creates AI startup incubator for employee alumni

• Alibaba sues Defense Department to be removed from blacklist

• Menlo Ventures raises $3 billion for AI firms in largest fund to date

• Seedance 2.5: The latest video model from ByteDance, an upgrade from its previous model with higher-resolution output and longer video duration.

• Beehiiv: The newsletter platform has added Cloudflare AI Crawl Control, which allows publishers to block or allow bots.  

• Mistral OCR 4: The model firm's latest model for document parsing, providing bounding boxes, content classification, and confidence scores. 

• Krea 2: The creative AI tool released the open weights for its models, called Krea 2 Raw and Krea 2 Turbo, undistilled models from mid-training meant to be fine-tuned.

• Sift: GTM Engineer

• xAI: Member of Technical Staff - Grok Product

• Google: Research Scientist, Visual Data and Generative Research

• IBM: AI Foundations - Research Scientist

GAMES

A QUICK POLL BEFORE YOU GO