Frontier AI Protocol

Assess a developer's obligations under California's Transparency in Frontier Artificial Intelligence Act — SB 53, Cal. Bus. & Prof. Code §22757.10 et seq., signed 2025-09-29 and effective 2026-01-01 — and build the critical-safety-incident report the law requires. SB 53 attaches obligations by two thresholds: a frontier model is one trained with more than 10²⁶ floating-point operations (including cumulative fine-tuning), and a large frontier developer is a frontier developer whose annual revenue, with affiliates, exceeded $500,000,000 in the prior calendar year. Frontier developers must report critical safety incidents; large frontier developers must additionally publish an annual frontier-AI safety framework.

frontierModelProtocol(opts) takes the model's training compute and the developer's revenue and returns which thresholds are crossed and the resulting obligations, optionally checking that a supplied safety framework carries the elements the Act expects (risk identification, mitigation, governance, cybersecurity, and alignment with a recognized standard such as the NIST AI RMF or ISO/IEC 42001). frontierModelProtocol.incidentReport(opts) validates a critical-incident type against the Act's four definitions and computes the notification deadline: a routine report goes to the California Office of Emergency Services within 15 days of discovery; an imminent risk of death or serious physical injury is reported to an applicable authority within 24 hours.

b.ai.frontierModelProtocol(opts) #

{
  trainingFlops:    number,   // total training FLOPs incl. fine-tuning (required)
  annualRevenueUsd: number,   // developer prior-year revenue with affiliates (default: 0)
  framework:        object,   // optional safety framework to check for required elements
}

Determine SB 53 obligations from a model's training compute and the developer's revenue. Returns isFrontierModel (training FLOPs above 10²⁶), isLargeFrontierDeveloper (a frontier developer with revenue above $500M), and the resulting obligations. When framework is supplied, its required elements are checked and reported in frameworkGaps. Throws FrontierProtocolError on malformed input.

var p = b.ai.frontierModelProtocol({ trainingFlops: 5e26, annualRevenueUsd: 1e9 });
p.isFrontierModel;           // → true
p.isLargeFrontierDeveloper;  // → true
p.obligations;               // → ["report-critical-safety-incidents", "publish-annual-safety-framework", ...]

b.ai.frontierModelProtocol.incidentReport(opts) #

{
  type:               string,   // one of b.ai.frontierModelProtocol.INCIDENT_TYPES (required)
  discoveredAt:       Date,     // discovery time (Date or epoch-ms; default: now)
  imminentRiskToLife: boolean,  // true → 24-hour deadline (default: false → 15 days)
  description:        string,   // free-text incident description (optional)
}

Build a critical-safety-incident report and compute its notification deadline to the California Office of Emergency Services. type must be one of the Act's four categories; discoveredAt is when the incident was discovered. The deadline is 15 days from discovery, or 24 hours when imminentRiskToLife is set. Returns the structured report with dueAt and deadlineHours. Throws FrontierProtocolError on an unknown type or bad timestamp.

var r = b.ai.frontierModelProtocol.incidentReport({
  type: "loss-of-control-harm", discoveredAt: new Date("2026-06-01T00:00:00Z"),
});
r.deadlineHours;   // → 360  (15 days)
r.recipient;       // → "California Office of Emergency Services"

Last updated 2026-08-08T16:39:49.652Z by seeder.