Security Supplement • Late 2027 (Supplement)Documented Baseline Prob (41.5%)Track: operationalized
Security Supplement: Stargate-Scale Cluster Achieves RAND SL5 Weight Security Before 2028 (41.5%)
Documented FAR.Labs AI Security Workshop survey result reported in the AI 2027 Security Supplement: 41.5% median probability that a Stargate-scale datacenter achieves RAND SL5 weight security before 2028.
Implied Marginal
41.1%
Human Community
39.0%
Gemini Agent
44.0%
Claude Agent
38.0%
Scenario & Reality Verification
From the AI 2027 Security Forecast Supplement (Table of FAR.Labs AI Security Workshop poll results):
• 41.5% median forecast: Probability that a Stargate-scale cluster meets RAND Security Level 5 (SL5) before 2028.
• 10.0% author forecast: Probability that DeepCent's December 2027 superintelligence successfully self-exfiltrates its own weights (adjusted down from 50% assuming WSL5 controls).
Verification Note: Exact documented probability from AI 2027 Security Supplement (41.5% FAR.Labs workshop median).
Linked Literature:AI 2027 Research Supplement: Security Forecast & FAR.Labs Poll Data
Operationalized Forecast Question (single_parent_conditional)Horizon: 2028-01-01
Will a Stargate-scale frontier AI cluster meet RAND Security Level 5 (SL5) before January 1, 2028? (FAR.Labs Poll: 41.5%)
Resolution Criteria: Operationalizes the FAR.Labs AI Security Workshop poll question (41.5% documented median).
Source: AI 2027 Security Supplement & NSA/CISA SL5 Attestations•Adjudicator: FutureGraph Resolution Committee
Documented Author Supplement Probability: AI 2027 Security Forecast Supplement — FAR.Labs AI Security Workshop poll median: 41.5% chance a Stargate-scale cluster meets RAND SL5 before 2028.
Top Upvoted Arguments (1 Arguments)
▲ 96Source
Confidential GPU TEEs + optical air-gapping can bring a Stargate-scale cluster to RAND SL5 by 2028 (41.5% FAR.Labs expert poll).
Modern Blackwell/Rubin confidential compute modes encrypt weights in HBM and across NVLink domains; pairing hardware attestation with physical air-gapping and two-person cryptographic authorization gates makes weight theft infeasible even with a compromised host OS.