Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallNew York Gov. Kathy Hochul signed the Responsible AI Safety and Education Act (RAISE Act) on December 19, 2025. It is a targeted frontier-model safety and transparency law—not a rule covering every chatbot, business software product or AI user. The signed law requires covered developers to publish safety practices, evaluate catastrophic risks, report qualifying critical incidents and follow their own commitments. A later amendment package would substantially revise the framework and set a January 1, 2027 effective date, but that package was listed as active rather than enacted on the New York Senate bill page available August 18, 2026.
What Hochul signed
The legislation, identified principally as S6953B/A6453B, is formally the Responsible AI Safety and Education Act. Hochul’s December 19, 2025 announcement described it as a nation-leading measure requiring safety frameworks and transparency for frontier AI models. State Sen. Andrew Gounardes and Assemblymember Alex Bores sponsored the bill.
Contemporary reporting said Hochul sought changes after lawmakers passed the measure and that the legislature agreed to address them through a subsequent amendment process. That history matters: the December signing and the later amendment package are separate events, and their provisions should not be merged.
Which AI developers are in scope?
The RAISE Act is aimed at developers of very large foundation models. It is not a general-purpose law for every generative-AI application.
Frontier-model threshold in the amendment text
The later S8828/A9449 amendment text defines a frontier model as a foundation model trained with more than 1026 integer or floating-point operations, counting the original training and later fine-tuning, reinforcement learning or other material modifications. Because this definition appears in the later amendment text, it should not automatically be treated as the operative definition of the December 2025 enactment.
#1 Best Overall
Large-developer threshold in the amendment text
S8828 defines a “large frontier developer” as a frontier developer whose annual gross revenue, including affiliates, exceeded $500 million in the preceding calendar year. That $500 million test is likewise a provision of the amendment text whose enactment status must be confirmed before relying on it as current law.
New York connection
The amendment text applies when covered models are developed, deployed or operated in whole or in part in New York. A company headquartered elsewhere therefore cannot assume that New York activity is irrelevant. Serving New York users alone should not be casually equated with the statutory nexus; the precise development, deployment or operation language controls.
Specified exemptions
The later text exempts accredited New York colleges and universities conducting academic AI-model research, and the Empire AI Consortium and associated institute. These are activity-specific exemptions, not a blanket exemption from every other New York AI requirement.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesWhat covered developers must publish and do
The framework is more than a one-time “safety report.” It combines governance, technical controls, public disclosures and incident response.
- Publish a frontier AI framework: The framework must explain how the developer incorporates national, international and industry safety standards; defines catastrophic-risk thresholds; assesses and mitigates those risks; reviews assessments before deployment or extensive internal use; uses third-party evaluators; updates the framework; protects unreleased model weights; responds to critical safety incidents; and governs internal use, including attempts to circumvent oversight.
- Review it at least annually: A material modification must be published with a justification within 30 days.
- Issue model-transparency information: Required information includes the developer’s website and a natural-person contact method, release date, supported languages and output modalities, intended uses, general restrictions, summaries of catastrophic-risk assessments, third-party evaluation involvement and other compliance steps.
- Follow the published framework: Failing to comply with a developer’s own safety commitments can itself be an enforcement violation.
The statute is best understood as a disclosure and governance regime backed by enforcement. It does not establish a state certification process approving every model, and it does not ban frontier-model development or deployment.
What counts as catastrophic risk?
Under the later amendment text, catastrophic risk means a foreseeable and material risk that a developer’s development, storage, use or deployment of a frontier model will materially contribute to a single incident causing either the death or serious injury of more than 50 people, or more than $1 billion in property damage or loss.
The qualifying scenario must involve a model that:
- Provides expert-level assistance to create or release a chemical, biological, radiological or nuclear weapon;
- Conducts a cyberattack or criminal conduct such as murder, assault, extortion or theft without meaningful human oversight; or
- Evades the control of its developer or user.
The definition excludes some situations, including information already publicly available in substantially similar form and harm to which the frontier model did not materially contribute. Terms such as “materially contribute” and “meaningful human oversight” remain interpretation questions for regulators and courts.
Recommended Free Tools
When must an incident be reported?
Critical-safety categories
The amendment text identifies four main categories:
Rank #3
- Unauthorized access to, modification of or exfiltration of model weights that results in death or bodily injury.
- Harm resulting from a catastrophic risk materializing.
- Loss of control of a frontier model that causes death or bodily injury.
- Deceptive model behavior that subverts developer controls or monitoring outside a designed evaluation and materially increases catastrophic risk.
72-hour rule
A covered developer must report to the oversight office within 72 hours after determining that a critical safety incident occurred, or after learning facts sufficient to establish a reasonable belief that one occurred. The clock is therefore tied to the determination or reasonable belief—not automatically to the moment the underlying event happened.
24-hour emergency disclosure
If an incident creates an imminent risk of death or serious physical injury, the developer must disclose it within 24 hours to an appropriate authority, such as law enforcement or a public-safety agency.
Reports are not necessarily public
The office may review submissions from developers and members of the public and transmit them to agencies including the attorney general. The amendment text exempts incident reports and catastrophic-risk assessments from New York’s public-records disclosure law, while also protecting trade secrets, cybersecurity information, public safety and national security. Accountability therefore does not mean that raw incident reports will appear in a public database.
Who enforces the RAISE Act?
Department of Financial Services office
The law establishes an office within the New York State Department of Financial Services, reporting to the superintendent. Its responsibilities include incident-reporting mechanisms, oversight, rulemaking, developer disclosures and annual reporting. The amendment text would require an anonymized, aggregated report beginning January 1, 2028, and annually afterward; that date belongs to the proposed amendment language unless and until the package is confirmed enacted.
Attorney-general penalties
The attorney general may bring a civil action for up to $1 million for a first violation and up to $3 million for each subsequent violation. Covered violations include failing to publish or transmit required documents, making prohibited or false statements, failing to report an incident or failing to follow the developer’s own framework.
The article creates no private right of action. Individuals cannot sue under the RAISE Act itself; someone harmed by an AI system would need to consider other state or federal legal theories.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.The amendment and effective-date question
The most important timing distinction is between the signed law and S8828/A9449. The New York Senate bill page listed S8828 as “ACTIVE” as of August 18, 2026 and reported that it was delivered to the governor on March 20, 2026. Its text would replace or revise parts of the original framework and set the effective date at January 1, 2027, replacing the original “90 days after becoming law” language.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Accordingly, the precise statement is: Hochul signed the RAISE Act in December 2025, while a subsequent chapter-amendment package would revise the regime and establish January 1, 2027 as its effective date. Publication should verify whether the amendments were enacted before treating that date, the $500 million revenue test or other S8828 provisions as operative law.
Why the law matters to companies and investors
For a covered developer, compliance is an ongoing operating function rather than a filing exercise. It may require documented risk thresholds, independent evaluations, model-weight security, governance approvals, public transparency reports, incident-response playbooks and rapid escalation to legal, technical and public-safety teams.
For businesses that buy or integrate AI, the law could make a developer’s published commitments and incident history important diligence material. It does not, however, impose the same obligations on ordinary enterprise users or every software vendor. Companies should first determine whether their own model-development activity, New York operations and corporate affiliations meet the operative statutory definitions.
New York’s approach also contributes to a patchwork of state frontier-AI rules. Developers may face overlapping disclosure and safety requirements, while policymakers debate whether common standards or federal legislation should replace state-by-state obligations.
Unsettled legal and practical issues
- Interpretation: “Reasonable belief,” “materially contribute,” “deceptive techniques,” “substantially modified version” and “meaningful human oversight” will need guidance, enforcement practice or litigation.
- Self-attestation: Developers generally publish and police their own frameworks. Regulators must decide how rigorous a framework must be and how to prove noncompliance.
- Confidentiality: The law seeks public transparency while shielding sensitive safety, security and commercial information.
- Federal conflict: Commentators have identified possible federal-preemption, Dormant Commerce Clause and First Amendment challenges involving compelled disclosures or interference with national AI policy. Those are potential theories, not established court outcomes. TechCrunch’s contemporaneous report provides that political and industry context.
- Implementation: The operation of the DFS office, final rules and the status of S8828 should be checked before making a compliance decision.
Bottom line for readers
New York did enact a frontier-AI safety law when Hochul signed the RAISE Act on December 19, 2025. Its practical focus is narrow but consequential: large developers of high-compute foundation models must document safety systems, publish specified information, report defined critical incidents and face attorney-general penalties for violations. It is not a universal AI law, not a consumer lawsuit statute and not a ban on advanced models. The later S8828/A9449 package is essential context, but its proposed January 1, 2027 effective date and revised thresholds should be treated as pending until its enactment is confirmed.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
