The Morning Download: OpenAI Releases Cybersecurity-Focused AI Model

Good morning. OpenAI is rolling out a cybersecurity-focused AI model in the latest sign that finding software vulnerabilities has become the industry's newest proving ground.

The move follows Anthropic's release last week of Claude Mythos, its own cybersecurity-focused model. The company decided to limit previews to a select group of organizations, including Amazon.com and Microsoft. 

OpenAI will offer the model, GPT-5.4-Cyber, to some participants of OpenAI’s Trusted Access for Cyber program, which the company rolled out in February, Bloomberg reports.  But the company in a blog post made a point of saying that it wants to make its cyber tools more widely available.

"We don't think it's practical or appropriate to centrally decide who gets to defend themselves. Instead, we aim to enable as many legitimate defenders as possible, with access grounded in verification, trust signals, and accountability."

All about the access. The emergence of AI models capable of spotting — and potentially exploiting — software vulnerabilities has put the question of access front and center: who gets these tools, and on what terms? Lock it down and defenders get outgunned. Open it too wide and attackers might have an easier time accessing a new weapon.

The debate is still being sorted out. Bloomberg reports that Treasury Chief Information Officer Sam Corcos was aiming to gain access to Mythos as soon as this week.

As AI gets smarter, catching its mistakes gets harder. AI chatbots may no longer instruct users to put glue on their pizzas, but hallucinations still happen. If anything, today’s more powerful chatbots may be more dangerous. Why? Millions have come to rely on (and trust) them as everyday tools.  And as Okahu CEO Pratik Verma tells the Journal, AI systems that produce more accurate results can be more likely to deceive users with inaccurate ones.

“When something is consistently wrong, the good thing is you know not to trust it,” he said. “But when things are mostly right but sometimes wrong, that’s the most pernicious one.”

Tired of it all? Your next stop should be Your AI Slop Bores Me. Chatbot answers delivered by people pretending to be chatbots.

Maine lawmakers pass ban on large data centers. Maine became the first state to ban large data-center construction Tuesday, as communities across the U.S. push back against the AI boom's toll on energy costs and the environment.

The construction freeze applies to data centers with capacities of at least 20 megawatts, which can power more than 15,000 homes, the WSJ reports. Democratic Gov. Janet Mills has said she would be in favor of a freeze if it includes an exception for a project already planned in the town of Jay.

For the AI industry and the businesses banking on it, this is more than a PR problem. The plot of the great AI infrastructure buildout until recently has focused on the race among tech giants to acquire capital, land, and GPUs to fuel their ambitions — addressing a capacity crunch that, the WSJ reports, hasn't eased. But alongside that scramble, a new set of forces is emerging that the industry can no longer ignore.

The grievances are concrete and personal. Maine has some of the country’s highest residential electricity prices and elected officials are concerned that a surge in data-center power demand might further inflate costs. At least 10 other states are advancing similar bills.

And in Mississippi... The NAACP is suing Elon Musk’s xAI alleging the company’s efforts to power its data centers in Tennessee and Mississippi are creating a health risk for local residents, the Journal reports. According to the lawsuit, the AI company and a subsidiary are illegally operating gas turbines without an air permit at a data center in Southaven, Miss.

Anthropic appoints Novartis CEO to board. Vas Narasimhan is the startup’s second new board addition in recent months as it eyes an initial public offering. The Claude developer added Chris Liddell, the former Microsoft and General Motors executive, to its board in February.

The company is also preparing its next flagship model, Claude Opus 4.7, along with a new AI-powered tool for designing websites and presentations, according to a person with knowledge of the products. Those new products could be released as soon as this week, the person told The Information.

Wegovy maker Novo Nordisk strikes deal with OpenAI to speed up drug discovery. The Danish drugmaker said it would integrate OpenAI’s models across its operations to help its workforce analyze complex datasets and reduce the time it takes to move from research to delivering treatments to patients, WSJ reports.

Sam Altman attack suspect charged with attempted murder. At the time of his arrest for a Molotov cocktail-style attack on the OpenAI’s CEO’s San Francisco home, 20-year-old Daniel Moreno-Gama was carrying an “Anti-AI” document that included the names and addresses of apparent board members and chief executives of AI companies, as well as investors, according to a federal criminal complaint filed Monday.

Amazon snaps up Starlink rival. Amazon.com is buying satellite operator Globalstar, which operates a network of satellites supporting Apple’s satellite connection feature on iPhones.
The companies estimated the deal at about $10.8 billion.

Rivian’s Illinois factory will run on recycled EV batteries. Electric-vehicle startup Rivian has found an unusual power source for its Illinois car factory: old batteries from its own cars. Rivian is joining with Redwood Materials to reuse EV batteries for energy storage—the largest repurposed-battery energy storage system for an automotive manufacturer in the U.S., executives told The Wall Street Journal.

UPS seeks to replace manual scans with tracking tech. United Parcel Service is rolling out technology to more closely track the billions of small packages that move through its U.S. network each year. The parcel-shipping giant said Tuesday it has invested $100 million to date to set up radio frequency identification, or RFID, technology across its network, with plans to spend more, WSJ reports.

A fallback plan to ensure Europe can defend itself using NATO’s existing military structures if the U.S. departs is gaining traction after getting buy-in from Germany, a long-term opponent of a go-it-alone approach. (WSJ)

The U.S. and Israel launched the war with the hope that killing top Iranian officials would create the conditions for regime change or at least the emergence of leaders more willing to bend to America and Israel’s interests. Instead, the void is being filled by radical new leaders who have shown little interest in political compromise at home or abroad. (WSJ)

Prosecutors from U.S. Attorney Jeanine Pirro’s office showed up unannounced Tuesday at the construction site for the Federal Reserve’s headquarters renovation, a new provocation in their investigation of the central bank. (WSJ)

The WSJ Tech Council brings together CIOs, CTOs and CISOs advancing innovation and shaping the future. Join this trusted community where tech executives connect with peers to explore emerging trends and gain the perspective they need to stay ahead of disruption.

Request Information

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.