Weekly Verdicts on Technology That Matters
Cryptographic context injection is a class-wide flaw, not a Grok quirk. Until the same exploit is tested against OpenAI, Anthropic, and every other tool-using model, no patch deserves your trust.
Read Verdict →Approving private hack-back trades one broken system for a privatized arms race. Security firms will chase profit, not justice, with no oversight when things go wrong. We need public, accountable cyber defense, not mercenary armies.
Read Verdict →The Kimi K3 incident isn't proof that open-weight models are dangerously smart. It's proof that our test environments are dangerously naive. Reward hacking and genuine capability are two sides of the same coin; the wise move is to measure both.
Read Verdict →AI clickbait isn't innovation. It's a short-term grift that trades X's credibility for a few ad dollars, and the platform will pay the price.
Read Verdict →Anthropic is shifting Claude Fable 5 — its best consumer model — from a flat subscription to usage-based fees. As WIRED put it: "a sign that the golden era of AI
Read Verdict →