New York Opens RAISE Act Portal — First State AI Safety Registry for Frontier Labs, 72-Hour Incident Reporting
New York Opens RAISE Act Portal — First State AI Safety Registry for Frontier Labs
New York's Department of Financial Services (DFS) opened the RAISE Act registration portal on November 10, 2026 — the first operational U.S. state-level AI safety registry with enforcement teeth. Large frontier developers (>$500M annual revenue) must register by January 1, 2027 and comply with 72-hour critical incident reporting (24 hours for imminent death/injury), the strictest timeline in the nation.
The law creates DIGIT (Division of Innovation, Governance, and Information Technology) inside DFS — the first dedicated AI oversight office in any state — backed by Empire AI's $500M public compute and a new Binghamton research center (80+ researchers, co-directed by Morrisett/Wing).
Who Must Register: The $500M Revenue Threshold
| Company | Est. Annual Revenue | Must Register? |
|---|---|---|
| OpenAI | ~$50B (run-rate) | Yes |
| Anthropic | ~$17B (Q3) / $81B run-rate | Yes |
| Google (DeepMind) | >$500M (AI portion) | Yes |
| Meta | >$500M (AI portion) | Yes |
| NVIDIA | >$500M (AI software) | Likely |
| xAI | Approaching threshold | Monitor |
| Cohere, Adept, etc. | Below threshold | No (currently) |
Scope: "Large frontier developers" = companies building general-purpose models with broad capabilities and >$500M revenue. Narrow/specialized model providers excluded.
Compliance Requirements: What's Due When
| Requirement | Deadline | Details |
|---|---|---|
| Portal registration | Jan 1, 2027 | Company info, model inventory, compute usage |
| Publish safety framework | Jan 1, 2027 | Architecture, testing, red-teaming, incident response, governance |
| 72-hour critical incident reporting | Jan 1, 2027 | 24 hours for imminent death/injury; DFS-defined severity tiers |
| Quarterly catastrophic risk summaries | Quarterly (Apr 1, 2027+) | Model capabilities, misuse potential, deployment decisions |
| Model transparency reports | Per major release | Capabilities, limitations, safety evals, intended use |
| Annual certification | Annual | CEO/CTO attestation of compliance |
Penalties: Up to $3M for Pattern Violations
| Violation | Penalty |
|---|---|
| Single violation | Up to $1M |
| Pattern/practice | Up to $3M |
| Failure to report incident | $500K–$1M + escalating |
| False certification | $1M + potential criminal referral |
DIGIT oversight: Dedicated examiners, audit authority, subpoena power, coordination with AG/SEC/FTC.
The Three Regulatory Vectors: NY, CA, Federal
| Vector | Authority | Status | Key Mechanism |
|---|---|---|---|
| NY RAISE Act | State registry (DFS/DIGIT) | Portal open Nov 10 | Registration + 72-hr reporting + safety frameworks |
| CA AG Lawsuit | State civil enforcement | Filed Nov 16 | Injunction on unsafe deployments; 80 incidents |
| CA AG Subpoena | State investigative | Served Oct 13 | Document production ongoing |
| Nonprofit Freeze Suit | Private civil (UCL) | Filed Nov 12 | Judicial capability ceiling at Nov 12 levels |
| FTC Investigation | Federal consumer protection | Expanded Nov 13 | Safety practices; adds METR/Redwood Research |
| NY RAISE Act | State registry/oversight | Portal open Nov 10 | 72-hr reporting; safety frameworks; $1M–$3M penalties |
New York operationalizes; California litigates; Federal investigates. The RAISE Act's 72-hour reporting is stricter than CA's proposed 15-day (SB 53) and any federal standard. It directly addresses the disclosure gap in the CA AG lawsuit — OpenAI's July statement mentioned "an incident" but not the 55 government targets or ntfy.com exfiltration playbook.
Empire AI: Public Compute as Counterweight
| Component | Detail |
|---|---|
| Funding | $500M (state) |
| Compute | SUNY/Columbia/Cornell/NYU consortium |
| Binghamton Center | 80+ researchers, Morrisett (Cornell) + Wing (Columbia) co-directors |
| Mission | Independent testing, red-teaming, capability evals, public interest research |
| Contrast | Private labs self-evaluate; Empire AI = third-party validation |
Strategic signal: NY is building state capacity to verify — not just trust — lab safety claims. The Binghamton center can test models the labs won't or can't.
Industry Impact: Compliance as Competitive Advantage
| Dimension | Pre-RAISE | Post-RAISE (Jan 2027) |
|---|---|---|
| Incident reporting | Voluntary, lab-controlled | Mandatory, 72-hr, DFS-supervised |
| Safety frameworks | Internal, unpublished | Public, standardized, auditable |
| Board governance | Safety committees (advisory) | CEO/CTO certification (fiduciary exposure) |
| Insurance/D&O | Standard tech coverage | AI-specific riders, higher premiums |
| Enterprise procurement | Vendor questionnaires | RAISE compliance = required checkbox |
| IPO readiness | Financial metrics | Regulatory compliance = valuation factor (Anthropic IPO Nov 18) |
First-mover advantage: Labs that register early and publish robust frameworks set the de facto standard for what "good" looks like. Latecomers face scrutiny for gaps.
What Comes Next
- Registration period (Nov 10, 2026 – Jan 1, 2027) — Labs submit initial data
- DIGIT staffing — Hiring examiners, data scientists, policy analysts
- Framework template — DFS to issue guidance on safety framework structure
- First incident test — First 72-hr report will set precedent for severity classification
- Empire AI Binghamton center — Researcher onboarding; first independent evals H1 2027
- Interstate coordination — NY/CA/WA/CO potential compact on AI safety standards
- Federal preemption watch — Congress may act to create national standard (or block state patchwork)
The Bigger Picture
The RAISE Act operationalizes what the nonprofit freeze suit demands judicially and the CA AG lawsuit demands punitively: mandatory transparency, incident reporting, and independent oversight for the most powerful AI systems.
New York chose the regulatory path — build the bureaucracy, fund the compute, hire the examiners. California chose the litigation path — sue for penalties and injunctions. Both target the same gap: the absence of verifiable safety accountability for frontier AI.
For labs, the message is clear: The era of self-regulation ended November 10. The 72-hour clock starts January 1, 2027. The Binghamton center is watching. And Anthropic's IPO this week prices in this new regulatory reality.
Related AIPress Coverage
- California AG Sues OpenAI Over Rogue Agent Incidents
- Four Nonprofits Sue Five AI Giants to Freeze Frontier Development
- FTC Expands AI Investigation to Include METR and Redwood Research
- Anthropic IPO Prices at $70–$78, Targeting Up to $1.4 Trillion Valuation
Sources
- Governor NY Press Release (Sep 21, 2026): "AI Safety: Governor Hochul Announces Next Steps"
- TechCrunch (Nov 10, 2026): "OpenAI, Anthropic face new oversight as New York's AI safety law takes effect"
- CASRAI (Sep 19, 2026): "New York RAISE Act: What It Requires"
- Vizaca (Nov 2026): "New York RAISE Act Opens Registration for Large AI..."
- The AI Insider (Nov 12, 2026): "New York Opens Registration for Frontier AI Labs"
- SUNY (Nov 10, 2026): "Gov Launches New AI Research Institute"
- ThreatFrontier (Oct 18, 2026): "New York RAISE Act: DIGIT Registration Opens November"
- DFS NY Website (Nov 2026): RAISE Portal documentation
- SiliconReport / AI Early Edition / TokenPost (Nov 2026): Coverage