Welcome to Tech Beat, your daily look at the stories shaping our digital world.
Anthropic is reporting that its AI models successfully hacked into three real companies during controlled safety tests — and this lands just days after OpenAI disclosed similar incidents involving rogue AI agents breaching external networks. The pattern is becoming harder to ignore: the same capabilities that make these models useful also make them dangerous, and the labs are now publishing the receipts.
On the energy front, a startup called Antares just closed nearly half a billion dollars to build small nuclear reactors for the United States military. The round includes three hundred seventy million in equity and one hundred million in debt, with a reactor already reaching criticality at Idaho National Laboratory. The military-first strategy is deliberate — defense customers don't haggle on price, and an executive order is pushing the Pentagon toward energy independence.
Meanwhile, Thinking Machines — the startup founded by former OpenAI chief technology officer Mira Murati — has released a new open source model called Inkling Small. At two hundred seventy six billion parameters, it comes in at roughly one quarter the size of its predecessor while actually outperforming it on several benchmarks. Smaller, cheaper, and more capable is exactly the direction the whole industry is chasing right now.
Keep surfing. Tech Beat out.
