Hochul Enacts Toned-Down RAISE Act for AI Oversight - AI News Today Recency

Gov. Kathy Hochul signed a revised, toned-down version of New York’s Responsible AI Safety and Education (RAISE) Act that narrows some original enforcement provisions while keeping new safety, reporting and oversight requirements for large AI developers in place[1][5].

What Hochul’s version of the RAISE Act does and does not do Hochul’s enacted bill requires major AI model developers to adopt safety protocols, report *critical safety incidents*, and submit plans describing how they would mitigate catastrophic harms, but the final text is materially less stringent than the earlier legislative draft that lawmakers initially passed[1][3].[6] Legislative aides and reporting describe that the governor’s version incorporated language from California’s SB 53—creating a regulatory approach closer to that law—and removed much of the original RAISE Act’s tougher penalty structure and some prohibitions on certain deployments that appeared in the June legislative version[1][2].[1] Lawmakers had sought civil penalties of up to $10 million for a first violation and $30 million for repeat violations in an earlier RAISE Act draft; Hochul’s chapter amendments pared back those maximums and other high-stakes enforcement elements before signing[2][1].

How the compromise was reached and the political dynamics Negotiations were reportedly contentious, with the bill at one point appearing likely to be vetoed before the governor and Legislature reached a compromise late in the process[1].[1] Tech industry groups and large platform developers pushed for a national, consistent framework rather than state-by-state rules; they lobbied for looser state rules and praised California’s approach as a baseline, while advocates and grassroots groups pressed for stronger guardrails and urged the governor to keep the tougher legislative text[1][3].[1]

What the law requires in practice The enacted RAISE Act creates several concrete obligations for covered entities, including requirements to maintain safety and security policies, to document measures limiting misuse and catastrophic failure modes, and to report incidents that pose critical risks to state authorities[6][3].[6] The law targets the largest actors in the AI ecosystem—those with substantial model-development resources—by tying applicability to thresholds such as training-cost or model-capacity metrics rather than to general revenue-based tests used elsewhere, reflecting differences between the New York and California statutes[2][6].

Reaction: advocacy groups, lawmakers and industry Advocacy groups hailed the law as a victory for accountability and public safety, calling it “a key victory” over industry pressure and praising New York for establishing enforceable guardrails[3][4].[3] Some lawmakers and technologists, however, described Hochul’s amendments as a retreat from the stronger penalties and prohibitions they had originally approved—characterizing the final statute as a compromise that preserved oversight but diluted enforcement teeth[5][1].[5] Industry stakeholders welcomed a more consistent and less punitive state approach, arguing that predictability and alignment with California’s law will reduce regulatory fragmentation while still requiring safety measures[1][2].

What to watch next: implementation and enforcement The law establishes a state-level oversight role and reporting regime; enforcement will depend on regulatory guidance, funding to the overseeing office, and how aggressively state officials interpret the statute’s safety and incident-reporting requirements[6][1].[6] Key implementation questions include how New York will define “critical safety incidents,” what documentation developers must submit, how penalties will be applied in practice, and whether the state will coordinate enforcement or information-sharing with federal regulators and other states to reduce a patchwork of rules[1][2].

Frequently Asked Questions

What is the RAISE Act that Gov. Hochul signed? The RAISE Act (Responsible AI Safety and Education Act) is New York legislation requiring large AI developers to implement safety protocols, report critical safety incidents, and provide documentation about measures to prevent catastrophic harms; Hochul signed a revised, less stringent version of the bill into law after negotiations with legislators and stakeholders[6][3][1].

How is Hochul’s version different from the Legislature’s original bill? Hochul’s enactment incorporated language from California’s SB 53 and removed or reduced several of the more severe enforcement provisions that appeared in the Legislature’s June draft—most notably by paring back proposed maximum civil penalties and deleting or weakening some prohibitions and mandatory remedies[1][2][1].

Who must comply with the law? The law targets the largest AI developers—those meeting statutory thresholds tied to model-development scale (such as training-cost or capacity metrics) rather than simply general revenue—so applicability focuses on well-resourced organizations building frontier models[2][6].

What kinds of incidents must be reported? Covered entities must report “critical safety incidents” to the designated state office; the statute and forthcoming regulatory guidance will clarify the incident types, thresholds, and timelines for reporting[6][1].

Will this create conflicting state rules for AI companies? Hochul and industry advocates emphasized the desire for national consistency to avoid a patchwork of state laws, and the governor’s incorporation of California-like language reduces some divergence; nevertheless, differences in thresholds and specific obligations mean state-by-state variation may persist unless federal standards emerge[1][2].

What are the next steps for enforcement and oversight? Implementation hinges on the state office created or empowered by the statute issuing guidance, allocating resources, and coordinating with other agencies; how aggressively the state enforces reporting, safety-plan adequacy, and penalties will determine the law’s practical impact[6][1].