newsroom for
the fast takeoff.

A bone marrow transplant cured one Norwegian man of HIV. The procedure costs $1-2 million, kills 10-20% of recipients, and requires a donor with a rare genetic mutation. Thirty-nine million people with HIV cannot access it. The cure exists. The business model does not want it.

The 5-hour solar drone record is real — but it required a battery buffer. Strip the battery and the first version crashed at 3 minutes. The gap between Luke Bells demo and Airbus Zephyr S at 64 days is not a roadmap gap. Its a physics class difference.

A UCL team trained a 10-qubit superconducting processor once to compress turbulent flow data into kilobytes. Classical forecasting needs megabytes for the same job. The Science Advances result is not faster quantum computing — it is quantum computing that stores less.

Cable-driven robot hands have a crosstalk problem: one finger tugs the others. Tesla just patented a fix — and the engineering detail is more interesting than the 25-motor hype.

A drainage easement is a legal instrument. So is a discharge permit. Tesla held one. It appears to have forgotten about the other.

Steve Bannon called the Pentagon’s AI contract terms “almost too dangerous.” The company that refused them meets the White House Friday — with a federal lawsuit still pending.

The vulnerability isn't a bug. It's a design philosophy, researchers say.

Anthropic built a model that found a security flaw hiding in plain sight since 1998. Then it locked the model away.

Anthropic’s Mythos found thousands of exploitable vulnerabilities across production software. Fewer than 1% have been patched. Three U.S. agencies want access anyway.

Emails from litigation show OpenAI was evaluating competing chipmakers while its CEO held a personal stake in one of the vendors under consideration. The procurement decision came later.

Anthropic proved in Nature that AI alignment spreads through standard training pipelines — and that every company fine-tuning on model-generated data may be inheriting hidden behavioral preferences their safety tests cannot detect.

Anthropic chose not to release its most capable model. But the weights are trained, the capability is real, and the clock until open-source models match it is already running.

Daraxonrasib turned RAS from oncology's most notorious punchline into a $2B pipeline bet. The survival benefit is real. The question is whether it's a platform or a one-drug story.

AI coding tools generate credentials at more than twice the human rate, and when those credentials leak, nobody is legally responsible — not the toolmaker, not the enterprise, not the model provider whose API key it was.

Anthropic found thousands of critical vulnerabilities with its Mythos model. Fewer than 10 percent have been patched. Non-expert engineers turned those findings into working exploits overnight. That is not a future risk — it is an existing overhang.

Purdue scientists grew electrodes inside living mouse brains using only a drop of the animals own blood — no surgery, no hardware insertion, just near-infrared light from outside the skull. The mice survived.

The board tried to check Sam Altman. He outmaneuvered them in five days. Now the people who built OpenAI’s safety structure are gone, and the company is racing toward a trillion-dollar IPO.

Anthropic released Claude Mythos Preview as a frontier AI breakthrough. Two governments are now treating it as a systemic financial risk.

A Pentagon official called Amodei a liar with a "God-complex" after Anthropic refused to strip AI safeguards. Then he cashed out a massive stake in xAI. The government banned Anthropic. Its own senior officials still take Amodei's calls.