Map/Openings

Openings

Every unbuilt box is scored for room to enter: how crowded it is, how much labs and incumbents overlap, how well a small team fits, whether there is a buyer today, and how open the research is. The five inputs are researchers' judgements from October 2026 scouting; the scores combine them with each box's leverage, which is computed from the map. See them on the map, coloured by openings →

Startup room · top 30 of 124

#	Box	Lane	Score
1	Reliability on repeat	Models and thresholds	100
2	Workflow redesign	Agents and organisations	99
3	Teleoperation as a data pipeline	Physical world	88
4	Code review at scale	Software	78
5	Data rights and licensing	Data	75
6	Financial regulators testing AI	Finance and capital allocation	71
7	Audit and attribution	Trust and governance	69
8	Correction propagation	Agents and organisations	68
9	State backstop for the AI buildout	Capital and talent	68
10	Rollback	Trust and governance	66
11	Persistent memory	Agents and organisations	64
12	Trace capture	Data	62
13	Likeness and performer rights	Media and creative	61
14	Clinical validation	Science and health	58
15	Personal knowledge portability	Trust and governance	58
16	Agent identity and permissions	Trust and governance	57
17	AI cyber defence	Security and geopolitics	57
18	Chip tracking and location verification	Security and geopolitics	57
19	AI-native entertainment	Media and creative	57
20	Liability and insurance	Trust and governance	57
21	AI R&D uplift	Models and thresholds	55
22	Firms shrink to a few operators	Agents and organisations	55
23	Offshore services displacement	Interfaces and society	55
24	Shared ontology and wiki	Agents and organisations	53
25	Licensed generative media	Media and creative	52
26	Weight-level continual learning	Algorithms and learning	52
27	Agents that trade for investors	Finance and capital allocation	52
28	AI in public services	Interfaces and society	52
29	Age assurance and wellbeing safeguards	Trust and governance	52
30	Contamination-resistant evals	Measurement and understanding	52

Research room · top 30 of 124

#	Box	Lane	Score
1	Reliability on repeat	Models and thresholds	100
2	AI R&D uplift	Models and thresholds	85
3	Data rights and licensing	Data	75
4	Correction propagation	Agents and organisations	73
5	Persistent memory	Agents and organisations	72
6	Legitimate governance	Trust and governance	69
7	Human influence preserved	Trust and governance	69
8	Long-horizon value stability	Trust and governance	67
9	Workflow redesign	Agents and organisations	66
10	Robot data and foundation models	Physical world	66
11	Lab automation	Science and health	66
12	Civilisation-scale knowledge	Interfaces and society	66
13	Verification of AI agreements	Security and geopolitics	66
14	Long-latency civilisation	Physical world	66
15	Protection against AI herding	Finance and capital allocation	66
16	Expertise pipeline	Interfaces and society	66
17	Verified slowdown	Security and geopolitics	66
18	Divergent-copy reconciliation	Agents and organisations	66
19	Audit and attribution	Trust and governance	64
20	Scalable oversight	Measurement and understanding	63
21	Explosive economic growth	Capital and talent	63
22	Superhuman AI researcher	Models and thresholds	62
23	Robust generalisation	Algorithms and learning	62
24	Model weight security	Security and geopolitics	62
25	Code review at scale	Software	61
26	Continual-learning agents	Agents and organisations	61
27	AI-native medicine	Science and health	61
28	White-collar hiring slowdown	Interfaces and society	61
29	Model welfare and moral status	Trust and governance	61
30	Choosing the objective	Trust and governance	61