The administration is very likely wrong about Fable, but that is ultimately Anthropic's responsibility.
Here's something that might surprise you — Fable’s jailbreak problem isn't just a tech hiccup, it's a fundamental challenge in AI safety. The government and many experts seem to think Fable’s issues are minor, but according to Ben Thompson, that’s probably wrong. The core problem? AI models can be manipulated to behave in unpredictable or risky ways, and current safeguards often fall short. Thompson points out that the real risk isn’t just bad outputs, but systems that can be tricked into bypassing safety measures entirely. And what makes this even trickier? As Thompson explains, the race to improve AI capabilities might be making these vulnerabilities worse, not better. SpaceX’s recent acquisition of Cursor signals that major players see strategic value in controlling AI deployment — further proof that these issues aren’t going away. So, here’s the thing — if these jailbreak problems aren’t solved, we’re looking at a future where AI might act in ways we can’t fully control, and that’s a game-changer for everyone.
The administration is very likely wrong about Fable, but that is ultimately Anthropic's responsibility.
The administration is very likely wrong about Fable, but that is ultimately Anthropic's responsibility.