NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me https://t.co/f3Aj9TYxU4
Each World Cup brings some interesting flights, but Santiago de Compostela to Nashville has to be near the top of the all-time list. https://t.co/vFupxlenyf
Meta is building dozens of massive tents at campuses across the US, sticking billions of dollars of chips inside, and powering them with off-grid turbines.
The AI race has officially entered its Mad Max phase.
Over the last month, I reviewed hundreds of documents and satellite images for Cleanview's latest report on behind-the-meter data centers. Meta's data center strategy, which is very visible from space, was one of the weirder approaches I came across.
Mark Zuckerberg recently ditched the data center designs that Meta had perfected over the last decade and told his team to stick tens of thousands of chips in tents outside their data center in New Albany, Ohio. Each of these chips costs about $60,000. Zuckerberg plans to stick billions of dollars worth of them in the tents.
The strategy has helped cut the time to build compute in half. The first five buildings at Meta’s New Albany, Ohio data center took between two and three years to build. Meta started building five ~125,000 square foot tents between April and June of 2026, according to city permits. Satellite images show the structures have all been built.
To power those "rapid deployment structures", as they are officially named, Meta signed a 10-year deal with Williams to build a pair of 200 MW off-grid power plants. Those power plants began construction about a year ago and are nearly complete.
Meta is using the same strategy to build a data center in Tennessee, bringing the total count of tent data centers to three.
Strategies like this are part of the reason behind-the-meter data center capacity is growing so quickly.
In Cleanview's report, I found that there's currently about 2 GW of BTM capacity online today. By the end of the year, it will likely be 3 GW—equivalent to three nuclear power plants. By the end of 2027, it could be as high as 13 GW—more than the power demand of NYC.
I've been talking to a lot of reporters about this research. When I told one reporter about these tents and other companies powering their data centers with jet engines, he said, "It's like a scene out of the movie Mad Max."
MAI-Thinking-1: el primer modelo fundacional de Microsoft SuperIntelligence (MSI). El equipo de MSI Madrid ha participado activamente en su diseño y entrenamiento! Y buscamos más AI engineers excepcionales para crear los próximos modelos de Microsoft desde España! DM para+info 🚀
Original Article: Daraxonrasib or Chemotherapy in Previously Treated Metastatic Pancreatic Cancer (phase 3 RASolute 302 trial) https://t.co/y4G27hfORg
#ASCO26 | @ASCO
Cheers, chills, and a standing ovation when RASolute 302 showed unprecedented survival on daraxonrasib for patients with progressive pancreatic cancer
Seldom do you sense you’re witnessing a historic moment in cancer care but this feels like ras targeting has arrived
#ASCO26
Eli Lilly has done it.
They've gone and made what seems to be a powerful, permanent gene therapy for LDL cholesterol.
That means they'll be able to effectively prevent most heart disease with a single infusion!
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.