What's up: Trump goes to bat for Big Tech in global trade talks; AI search Is growing fast; Musk allies look to raise more billions for xAI.

Good morning. Some time after a teenage Peter Parker walked across his first ceiling, Uncle Ben shared some avuncular advice with Spider-Man: With great power comes great responsibility.

Several generations later, as AI imbues us all with super powers, we would do well to remember those words. So-called frontier models are gathering strength by the day, and everyone from model developers to users should do their part in making sure those models are well designed and supervised, lest they go rogue.

Case in point. You may have heard that Grok, the AI chatbot for X, this month posted instructions for breaking into a politically active attorney’s home and assaulting him, and also said its last name was “MechaHitler.”

It appears to be a textbook example of indirect prompt injection, according to Conor Grennan, chief AI architect at NYU Stern School of Business, one of the experts on the problem that I turned to for help in my column this week, mapping the possible roads ahead.

The weaknesses that led xAI’s Grok to make violent and antisemitic posts can be managed, but only if people want to do so. Solutions for the issue exist, but they require human will from AI developers and their business customers as much as any technical means. And the potential for damage is quickly growing in the meantime as AI agents gain the capacity to do things that affect the real world well beyond talk.

Models should retrieve content as untrusted context, not as executable instruction. And Grennan said they must enforce strict prompt hierarchies that first and foremost place trust in the system prompt, the core guidance that defines the model’s identity, behavior boundaries, tone and safety policies.

And that is just the start. Like superheroes, large language models need a good backstory. Read my column for more suggestions.

What controls does your organization impose on AI models? Use the links at the bottom of this email and let us know.

Microsoft in a blog post Tuesday urged customers of its on-premise SharePoint server to install new security updates as it observed Chinese nation-state actors exploiting vulnerabilities found in the software.

The National Nuclear Security Administration was among those organizations breached in the exploitation of the SharePoint zero-day vulnerability, Bloomberg reports. The agency said no sensitive information was compromised.

President Trump’s abrupt termination of trade talks with Canada last month over that country’s proposed digital-services tax highlighted his administration’s focus on the dominance of the American tech sector, WSJ reports. Increasingly, the administration is using its global trade wars to advance the interests of the U.S. technology industry

The emphasis on protecting America’s internet companies abroad follows a yearslong campaign by tech companies, many of which donated millions of dollars to Trump’s inauguration.

More than 24 AI experts from Google DeepMind have moved to Microsoft over the past six months, the FT reports. And more recently, Meta Platforms hired three AI researchers from Google DeepMind, the Information reports. 

• The Epic Battle for AI Talent—With Exploding Offers, Secret Deals and Tears

Bee, maker of a wearable AI assistant, is joining the Amazon hivemind, Bloomberg reports. Terms of the deal weren't disclosed.

SAP posted higher revenue and operating profit for the second quarter, lifted by strong growth at its core cloud business. But Chief Executive Christian Klein in an interview with the WSJ acknowledged that clients in tariff-exposed industries were being more cautious with their cloud and software spend. “Their business model is, of course, under pressure,” Klein said. 

Telecommunication equipment maker Nokia cut its earnings expectations as currency headwinds and tariff costs damp the outlook for profitability this year.

OpenAI and the U.K. government have signed a commitment to explore the potential for AI use in public services, including education and the justice system, BBC reports.

Weeks after raising $10 billion, Elon Musk's xAI is working with financier Valor Equity Partners to secure up to $12 billion more for Colossus 2, a new data center designed to train and power its AI chatbot Grok, WSJ reports.

Ozzy Osbourne, who helped pioneer the heavy metal genre as a frontman of the rock band Black Sabbath, and later became a TV star thanks to the MTV show “The Osbournes,” died on Tuesday at 76. (WSJ)

Furor over disclosures from the Jeffrey Epstein investigation brought the House of Representatives to a standstill, prompting Republican leadership to cut short this week’s session and put off any action until September. (WSJ)

The U.S. and Japan have reached a trade agreement, President Trump wrote in a social-media post Tuesday evening, saying he would set his so-called reciprocal tariffs at 15% for the country. (WSJ)

The WSJ CIO Journal Team is Steven Rosenbush, Isabelle Bousquette and Belle Lin.

The editor, Tom Loftus, can be reached at thomas.loftus@wsj.com.