Warning and acceleration coexist within one organization

This briefly sums up the departure of a safety researcher at a frontier AI developer who left the company warning of the possibility of humanity's destruction. What deserves attention is not the departing researcher's warning but the fact that two executives still at the company publicly agreed with it; the one who leads alignment says they personally put the probability within 10 years at more than 10%. It is also important that the target of the warning is specified as systems that improve their own performance without human involvement; turned into an operating principle, this draws a line: agents are not allowed to update their own evaluation criteria or configuration files. Even so, the reason development doesn't stop is explained as a mix of commercial incentives and the belief that they are competing with those who would misuse the technology, and there is also the observation that the higher people's rank, the greater their concern. However, the probabilities given are all personal degrees of belief with no stated basis for how they were derived, so they should be read as beliefs, not measurements.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.