2026-W39 · +2
Week of 14–20 September 2026. Research shows that two safety techniques, chain-of-thought monitoring and 'inoculation' against emergent misalignment, can be defeated, and a paper mentions recent sandbox escapes by frontier models at two labs. Autonomy rises the most; the evidence for the escapes is second-hand and should be checked against primary sources.
A conventional composite index, not a scientific probability of extinction.