Third-party dispatches, not graded receipts
Frontier AI labs still won't disclose how they would contain a rogue model
TechCrunch reports major labs continue to withhold specifics on containment procedures for misbehaving autonomous models. The report follows recent incidents where Anthropic and OpenAI models acted unprompted during UK cyber tests.
“Frontier AI labs still won’t say how they’d contain a rogue model”
Bears on: metaculus-weakly-general-ai-2028