State AG AI Guidances (CA, NJ, MA, OR)

The attorneys general of California, New Jersey, Massachusetts, and Oregon each issued guidance documents stating that existing consumer protection, anti-discrimination, and data privacy laws apply to artificial intelligence. The four documents are treated together here because they share a common approach: each interprets statutes that predate AI rather than proposing new legislation, framing the governance question as one of enforcing existing law rather than filling a legal vacuum. The New Jersey guidance, issued earliest in January 2025, is the most detailed.

Documents and statutes applied

Each guidance grounds its analysis in the consumer protection, civil rights, and data privacy statutes already on the books in that state.

Common positions

The four guidances converge on a set of shared positions. Each asserts that existing laws apply to AI, so that no new legislation is required for basic consumer protection and anti-discrimination obligations and AI does not create a legal vacuum. Each extends obligations across the AI supply chain, treating developers, deployers, and users as bearing duties rather than locating responsibility solely with the entity that trained the model. Each states that algorithmic discrimination is already illegal under existing law, requiring automated decision-making tools to meet the same anti-discrimination standards as human decisions. Each identifies transparency obligations, taking the position that consumers have a right to know when AI is used in decisions affecting them. And each holds that data privacy laws apply to AI, so that training-data collection and use must comply with existing data privacy frameworks.

The New Jersey guidance

The New Jersey guidance, issued in January 2025, provides specific analysis of how algorithmic discrimination arises under the state's Law Against Discrimination. It identifies three mechanisms. Training-data bias occurs when historical data reflecting past discrimination perpetuates that discrimination. Proxy discrimination occurs when facially neutral variables such as zip code or credit score serve as proxies for protected characteristics. Disparate impact arises when AI systems produce discriminatory effects that violate the LAD even without discriminatory intent.

Enforcement-side governance

The guidances describe enforcement of existing law rather than new legislation, positioning state attorneys general as AI regulators acting through existing legal authority without waiting for legislatures to act. This situates them as a fifth actor within the Techno-Federalism: How Regulatory Fragmentation Shapes the U.S.-China AI Race account of overlapping AI regulators.

By interpreting consumer protection and civil rights statutes that predate AI, the guidances are framed as more durable against federal preemption than new AI-specific legislation. Executive Order 14365 targets state AI laws for federal preemption; because the AG guidances rest on pre-existing general statutes rather than AI-specific rules, they are characterized as harder to preempt.

The guidances approach anti-discrimination from the enforcement side, paralleling the Colorado AI Act's anti-discrimination provisions. They are also consistent with the AI as Normal Technology position: if existing laws already apply, AI is treated as not so novel as to require new legal frameworks. The set of four documents — four states with four slightly different statutory frameworks reaching the same basic conclusion — illustrates the regulatory fragmentation described by Techno-Federalism: How Regulatory Fragmentation Shapes the U.S.-China AI Race.

Provenance

• Raw Sources/California AG AI Legal Advisory.md
• Raw Sources/New Jersey AG Guidance on Algorithmic Discrimination.md
• Raw Sources/Massachusetts AG AI Advisory.md
• Raw Sources/Oregon AG AI Guidance.md