Anthropic, Amazon, Microsoft, and Google Propose a Cross-Lab Framework to Score AI Jailbreak Severity

My take: This one matters, even if the headline sounds dry at first glance. Anthropic, together with Amazon, Microsoft, and Google, is proposing a shared four-axis framework to score how serious an AI jailbreak is — how much power the attacker gains, how broad the potential damage is, how easy the attack is to weaponize, and how widely known the technique already was. This came out of the June chaos when the U.S. government banned Fable 5 globally because a jailbreak was reported with no shared scale to evaluate how serious it actually was. It's like firefighters from four different countries showing up to the same fire, but each with their own idea of what counts as a 'major emergency' 🔥. With a shared rubric, the next time a jailbreak is reported, governments and companies speak the same language instead of making panic decisions. They're also launching a HackerOne bug-bounty program for security researchers. To me, that's real progress.