Altman's Calculated Gamble: The Thermodynamics of AI Risk vs. Reward

When OpenAI Chief Executive Sam Altman frames the trajectory of artificial general intelligence (AGI), he repeatedly invokes a stark formulation: the upside of intelligence too cheap to meter is virtually limitless, while the catastrophic failure modes—ranging from automated bioweapon synthesis to self-exfiltrating autonomous code—could prove existential. This is not mere rhetorical hyperbole; it is the fundamental optimization dilemma governing multi-hundred-billion-dollar frontier clusters.

Our governance matrix computes the frontier curve by balancing training compute scale (FLOPs) against alignment tax and agentic tool permissions. When scaling outpaces containment infrastructure, a model's latent autonomous capability develops faster than verification scaffolds can detect reward-hacking or deception. Modern governance frameworks, including the EU AI Act's General Purpose AI (GPAI) systemic risk designations and the US NIST AI Risk Management Framework, require verifiable safety margins before moving from closed red-teaming to open commercial distribution.

Digital Necromancy: California § 3344.1 and New York § 50-f in the Age of AI

The legal infrastructure governing the synthetic recreation of deceased performers is fragmented across state jurisdictions. Unlike copyright, which is federally preempted, right-of-publicity claims depend on the deceased domicile at death:

• California's Astaire Celebrity Image Protection Act (Civil Code § 3344.1): Protects a deceased personality's name, voice, signature, photograph, and likeness for 70 years post-mortem, requiring registration with the California Secretary of State to enforce commercial licensing claims.
• New York Civil Rights Law § 50-f (2020): Extends 40-year post-mortem publicity rights for deceased performers specifically targeting synthetic digital replicas in audiovisual works, with explicit exemptions for historical documentaries, parody, and satire.
• Federal NO FAKES Act & ELVIS Act (Tennessee): Emerging federal and state standards that classify an individual's unique voice and digital persona as an actionable property right, imposing strict liability on platforms distributing non-consensual digital replicas.

When studios synthesize historical icons like Greta Garbo or James Dean, estate consent alone may not shield producers from Lanham Act Section 43(a) false endorsement claims or consumer fraud actions if audiences are misled into believing historical statements were made by the living individual without clear cryptographic C2PA provenance.