Anthropic의 Fable 5 복귀 — 하지만 안전 필터가 무해한 코드까지 차단
Anthropic은 탈옥(jailbreak) 문제로 인한 2주간의 금지 조치 이후 Fable 5에 대한 글로벌 접근을 복원했지만, 새로운 안전 수정이 무해한 프로그래밍 작업까지 잘못 차단하고 있습니다.
anthropic lifted the suspension on fable 5 following a two-week government investigation into a vulnerability that allowed the model to generate software exploits. the fix involved training a classifier that blocks the specific jailbreak technique in over 99 percent of cases, though users will experience increased refusals on routine coding tasks. the less restricted mythos 5 remains limited to us government-approved partners, while anthropic admits making models impervious to jailbreaks is probably impossible. the company is establishing a 24/7 monitoring team and a hackerone program to handle future vulnerabilities, signaling a shift toward tighter government integration on frontier model safety.