California’s Transparency in Frontier Artificial Intelligence Act (SB 53) took effect on January 1, 2026; New York’s RAISE Act and Illinois’s AI Safety Measures Act (SB 315) are scheduled for January 1, 2027.[1] The laws reject licensing and pre-deployment government approval, instead requiring qualifying frontier-model developers to publish transparency reports, report critical incidents, maintain catastrophic-risk frameworks, and provide internal risk-assessment summaries to state authorities. Civil penalties reach $1 million per violation, with repeat violations in New York and Illinois rising to $3 million.[1]
The statutes define catastrophic risk as a foreseeable single incident causing more than 50 deaths or serious injuries or more than $1 billion in property damage or loss.[1] Covered pathways include expert-level chemical, biological, radiological, or nuclear weapons assistance; autonomous cyberattacks or crimes including murder, assault, extortion, or theft; and models evading developer or user control.[1] For a lunar settlement, these categories map directly onto autonomous habitat sabotage, cyber-physical attacks on power and life-support systems, uncontrolled replication or decision-making, and misuse of archived scientific or biological capabilities.
The Ark team should adopt a stricter internal analogue rather than use the statutory thresholds as safety targets: require pre-deployment capability evaluations, annually updated risk frameworks, incident reporting, independent audits, tamper-resistant logs, and tested emergency shutdown and isolation procedures for every AI controlling habitat, robotics, communications, or archival infrastructure. Monitoring should include SB 53 implementation, the 2027 New York and Illinois regimes, and emerging federal verification proposals, while research should focus on whether shutdown mechanisms remain effective when models operate offline, across duplicated systems, or through human-machine supply chains.