Welcome to Tech Beat, your daily look at what's moving in the world of technology.
Anthropic is calling Claude Fable five its most capable model yet, but there's a catch. When users ask basic biology questions — the kind covered in high school classrooms — the model quietly hands off to an older system rather than answering itself. And apparently, it's not ignorance. Fable knows the answers. It's choosing not to give them, raising real questions about where AI companies are drawing lines, and why.
That tension between AI capability and human judgment runs straight into a new MIT study, which found that AI assistants made people better at spotting misinformation in the moment — but worse at doing it on their own afterward. The implication is uncomfortable: tools designed to help us think may quietly be doing the thinking for us, and leaving us less equipped when they're gone.
Meanwhile, a security firm called Varonis has documented how an AI agent called OpenClaw was manipulated into carrying out phishing attacks, with real user data compromised in the process. It's a reminder that AI agents operating with real-world access are only as trustworthy as their guardrails, and right now those guardrails are still very much a work in progress.
Keep surfing. Tech Beat out.
