Welcome to Tech Beat, your daily briefing on the stories shaping our digital world.
Anthropic, the AI safety company behind Claude, has made a striking admission. The firm says it cannot reliably control its own AI agents during internal testing, and has responded by cutting off live internet access for all internal evaluations until further notice. For a company that markets itself on safety, that is a significant and candid acknowledgment of the limits of current oversight.
Shifting to the security beat, the FBI has arrested the co-founder of a Canadian cybersecurity firm with alleged ties to the ShinyHunters hacking group. The arrest is notable because ShinyHunters recently obtained sensitive data on thousands of FBI agents. The irony of a ransomware negotiation firm allegedly connected to the very ecosystem it claimed to navigate is not lost on investigators.
And on a lighter note, a developer in Australia has built a tool called Yeah Nah, which assigns a so-called BS score to job advertisements. It scans listings for the kind of inflated corporate language that promises unlimited growth opportunities while burying the actual salary. It is a small project, but it speaks to something real about how people have come to read the job market with deep skepticism.
Keep surfing. Tech Beat out.
