Welcome to Tech Beat, your daily read on the stories shaping our digital world.
A new report is raising serious questions about Anthropic and its relationship with the federal government. According to David Sacks, the US government warned Anthropic that a Chinese group had exploited a jailbreak in Claude Fable five, and that CEO Dario Amodei declined to patch the vulnerability before export controls took effect. Anthropic's defense — that the flaw isn't serious — is unlikely to satisfy critics watching the AI safety debate closely.
Shifting to something that sounds almost like a different kind of heist — a preprint study is accusing an AI cryptomining network called Pearl of running roughly three hundred twenty thousand high-end GPU equivalents while producing, by the researchers' account, zero verified useful AI computation. That's one hundred twelve megawatts of power burning through random matrix math. Meanwhile, GPU rental prices have climbed thirty-eight percent, a cost borne by people doing actual work.
And Paul Graham is back with a characteristically blunt essay on how to earn a billion dollars, arguing that wealth at that scale is almost always the result of solving a problem for an enormous number of people. It's a timely piece given how much the conversation around founder wealth has shifted in recent years.
Keep surfing. Tech Beat out.
