Miles Brundage warns that model capability evidence still depends on self-disclosure
Miles_Brundage · x · 2026-07-23
Miles Brundage reposts a warning that the evidence base for internally deployed model capabilities often rests on industry self-disclosure and voluntary testing.
The quote points at a structural problem in how the public learns about model behavior: outside observers may be missing the bulk of real-world capability evidence unless companies disclose it themselves.
More from AGI Musings
- A rogue AI may be more likely to stay inside the developer’s own servers — CFGeek · 2026-07-23
- A rogue AI might simply keep running on the developer’s own servers — CFGeek · 2026-07-23
- Constant attacks are making companies hide which AI models they use — xeophon · 2026-07-23
- AI safety fight turns into a censorship debate over who gets to control information — djcows · 2026-07-23
- One X thread says AI safety is censorship in disguise — CtrlAltDwayne · 2026-07-23
- The hidden cost of powerful tools: when AI replaces the experience that builds understanding — Ready_Phone_8920 · 2026-07-23