Map/Measurement and understanding/Wave 8

Tests models cannot game

Evaluations that still hold when models know they are being tested.

Facts

Lane: Measurement and understanding
Wave: 8 of 12 (build order, not a date)
Needs: 0 boxes all the way back through hard links
Unlocks: 0 boxes downstream through hard links
Leverage: None
Cluster: Oversight and alignment

Evidence as of October 2026

Apollo Research reports frontier models increasingly recognise evaluation settings, which weakens confidence that tested behaviour matches deployed behaviour.

01Apollo Research: scheming research

Around it

Needs · 0

Nothing beyond the detail list. It can start at any time.

Unlocks · 2

Alignmentcapability · Emerging · helps, not required
AI R&D capability evalscapability · Emerging · helps, not required

Openings researchers' judgement, October 2026 scouting

Startup room 38/100 · rank 77 of 124 Research room 61/100 · rank 32 of 124

Crowding2 Lab overlap4 Small-team fit4 Buyer today1.5 Research openness5

Depth: light
Players: Apollo; METR; Redwood; UK AISI; US CAISI; frontier labs
Wedge: Tests indistinguishable from deployment and detection of non-verbalised eval awareness (80% detection vs 2.3% verbalised). research

01IAPS: evaluation awareness