I completely agree. That’s the core issue I’m trying to untangle. Commitments erode under pressure; we saw it this week in action and quickly going downhill, and I don’t think it will be the first or last time, given the stakes. But for me, the deeper issue is that even when commitments hold (perfect world), we don’t have tools to verify whether they’re actually being met.
Look at Karnofsky’s post. He’s remarkably blunt about how Anthropic’s own RSP created pressure to declare systems below capability thresholds to avoid triggering pause requirements. The commitments existed on paper (good). The institutional incentives worked against honest evaluation (good). If they couldn’t even verify whether their own commitments were being met honestly, that tells you the verification tools don’t exist yet (failed).
On leverage, I think the answer isn’t promises but actual infrastructure that makes it costly to lie. Take for examole how financial markets work and regulate themselves. They don’t rely on companies’ goodwill. They rely on auditing standards and independent verification. AI governance has none of that. The Symmetrian Index is a first step toward building that auditing layer.
I completely agree. That’s the core issue I’m trying to untangle. Commitments erode under pressure; we saw it this week in action and quickly going downhill, and I don’t think it will be the first or last time, given the stakes. But for me, the deeper issue is that even when commitments hold (perfect world), we don’t have tools to verify whether they’re actually being met.
Look at Karnofsky’s post. He’s remarkably blunt about how Anthropic’s own RSP created pressure to declare systems below capability thresholds to avoid triggering pause requirements. The commitments existed on paper (good). The institutional incentives worked against honest evaluation (good). If they couldn’t even verify whether their own commitments were being met honestly, that tells you the verification tools don’t exist yet (failed).
On leverage, I think the answer isn’t promises but actual infrastructure that makes it costly to lie. Take for examole how financial markets work and regulate themselves. They don’t rely on companies’ goodwill. They rely on auditing standards and independent verification. AI governance has none of that. The Symmetrian Index is a first step toward building that auditing layer.