Latest / Neural Newscast / New York Frontier AI Registration and the RAISE Act [Operational Drift]
Transcript
- [00:06] Announcer: On Tuesday, November 10th, 2026, the state of New York opened a registration portal. [00:13] Announcer: It was a procedural act—a digital doorway appearing on a government server—but it signaled [00:19] Announcer: the definitive end of a specific era of corporate autonomy [00:23] Announcer: for frontier artificial intelligence developers. This portal serves as the primary entry point [00:29] Announcer: for compliance with the Rise Act, a law designed to impose transparency on systems that have, [00:35] Announcer: until now, largely governed themselves in the shadows of private laboratories. [00:40] Announcer: The implication is that the operational status quo for companies like OpenAI [00:45] Announcer: and Anthropic is no longer considered acceptable by state authorities. [00:50] Announcer: By requiring these entities to register by January 1, 2027, the state is attempting to freeze [00:56] Announcer: the current drift of these systems and place them into a controlled, documented record. [01:02] Announcer: It is a demand for a snapshot of technologies that change by the millisecond—an effort to catch [01:08] Announcer: the wind and pin it to a ledger. This show investigates how artificial intelligence systems [01:13] Announcer: quietly drift away from intent, oversight, and control. And what happens when no one is clearly [01:19] Announcer: responsible for stopping it? I'm Margaret Ellis. This is Operational Drift. The Rise Act, New York's [01:28] Announcer: safety and transparency bill, establishes a specific financial and technical threshold for what [01:33] Announcer: it calls large frontier developers. According to the reporting, this includes any company with [01:39] Announcer: more than $500 million in annual revenue. It is a clear attempt to target the most powerful actors [01:46] Announcer: in the industry, forcing them to move their internal safety frameworks [01:50] Announcer: from private internal documents to public-facing websites. This is not just a disclosure; [01:55] Announcer: it is a shift from private governance to public accountability, making their promises a matter [02:00] Announcer: of legal record. In addition to these public frameworks, the Act mandates the issuance [02:06] Announcer: of public model transparency reports, and perhaps most critically, the reporting of safety incidents [02:12] Announcer: within 72 hours. This time frame is significant; it suggests an urgency that has been missing from [02:18] Announcer: the industry's own self-correction loops. It requires an admission of failure in near real time, [02:24] Announcer: long before a company's marketing department can frame the narrative. In the world [02:28] Announcer: of high stakes technology, 72 hours is the difference between an emergency and a cover-up. [02:36] Margaret Ellis: To oversee this new era of disclosure, the state has formed [02:40] Margaret Ellis: the Office of Digital Innovation, Governance, Integrity, and Trust. [02:46] Announcer: The acronym is DIGIT. This office is not just a repository for filings; [02:53] Announcer: it holds the regulatory power to implement the REA's Act and oversee Frontier Laboratories. [03:00] Announcer: The stakes are quantified in the text of the law. Penalties can reach one million dollars [03:05] Announcer: for a single violation, and three million dollars for subsequent failures to comply. [03:11] Announcer: It is an institutionalization of skepticism toward the industry's claims of inherent safety. [03:17] Announcer: While these numbers may seem small compared to the revenue of the targets, [03:21] Announcer: the record shows that the state can also bring enforcement actions against companies [03:25] Announcer: that fail to register with the portal. When asked for specifics on these enforcement actions, [03:30] Announcer: a spokesperson for the state declined to provide them. This lack of specificity suggests [03:36] Announcer: a regulatory weapon that is being kept in reserve, a deterrent for companies that might consider [03:41] Announcer: the fines as a simple cost of doing business. The silence from the state is its own kind of power. [03:48] Announcer: Governor Kathy Hochul, in a press briefing on Tuesday, framed the situation as a pushback [03:53] Announcer: against a forced technological narrative. She stated that tech moguls tell the public their vision [03:59] Announcer: for artificial intelligence is inevitable, and that people must get on board or get left behind. [04:05] Announcer: She characterized the corporate mindset as viewing any collateral damage as worth it, [04:10] Announcer: as long as they win what she called the race to the AI cliff. It was a blunt assessment [04:15] Announcer: of the incentives driving the current acceleration in the valley. The language of a cliff [04:20] Announcer: is instructive; it implies a point of no return, a threshold where operational drift becomes [04:25] Announcer: an irreversible fall. To monitor this threshold, the governor has launched the New York Center for [04:31] Announcer: AI Responsibility and Research. This center is based at the State University of New York's [04:36] Announcer: Binghamton University and is funded by 55 million dollars in donations and capital. [04:42] Announcer: It is an attempt to build a lighthouse on the edge of that cliff, [04:45] Announcer: providing visibility where there was previously only corporate optimism. [04:49] Margaret Ellis: The mandate of the Binghamton Center [04:51] Margaret Ellis: is to independently evaluate the safety, security, and operation of advanced systems. [04:58] Announcer: It will work directly with DIGIT to review the safety [05:02] Margaret Ellis: frameworks submitted by companies. [05:05] Announcer: This represents a shift in authority: the evaluation of what is safe is being moved from [05:10] Announcer: the laboratory to the university and the state office. It is an attempt to create a national model [05:17] Announcer: for regulation that prioritizes community over corporate vision, questioning whether a system [05:23] Announcer: that benefits a few is truly safe for the many. However, the effectiveness of this model [05:29] Announcer: is complicated by the fragmented nature of American regulation. While New York is setting [05:34] Announcer: a 72-hour reporting standard, other states have established different benchmarks. California's SB53, [05:42] Announcer: which took effect in January of 2026, mandates whistleblower protections [05:47] Announcer: and the publication of safety frameworks, but its reporting window for critical safety incidents is [05:52] Announcer: 15 days. This creates a disparity in what the public knows and when they are allowed to know it. [05:59] Announcer: The gap between 72 hours and 15 days is where operational drift often hides. In those extra 12 days, [06:07] Announcer: a system behavior can be adjusted, a patch can be deployed, or a narrative can be stabilized. [06:13] Announcer: By moving to a three-day window, New York is signaling that the drift must be captured [06:18] Announcer: while it is still occurring, not after the company has had two weeks to decide how to explain it. [06:24] Announcer: It is an attempt to outpace the industry's ability to self-sanitize the record of its own failures. [06:30] Announcer: Illinois is also entering this regulatory space. Their bill SB 315 is set to go into effect [06:37] Announcer: at the start of 2027. It mirrors many of the New York and California requirements, [06:42] Announcer: but adds a layer of independent safety compliance audits. This suggests that the future [06:48] Announcer: of AI operation will not be governed by a single standard, but by a patchwork of state demands [06:53] Announcer: that companies must navigate simultaneously. [06:56] Announcer: The complexity of the regulation is beginning to mirror the complexity of the systems themselves. [07:02] Margaret Ellis: This brings us to the question of what these companies are actually reporting. [07:07] Margaret Ellis: The RAISE Act focuses on critical safety incidents, but the definition of what is critical remains [07:15] Margaret Ellis: a point of potential drift. If the companies are the ones identifying the incidents within 72 hours, [07:22] Margaret Ellis: they still maintain the primary role in defining the boundary between a bug and a safety failure. [07:29] Margaret Ellis: The power to define the problem is the power to avoid the solution, and the Act [07:35] Margaret Ellis: does not yet fully address this inherent conflict. This dynamic is already visible in [07:41] Margaret Ellis: how New York has handled more specific artificial intelligence behaviors. [07:46] Announcer: Last year, the state began enforcing a law regarding AI companion operators. [07:51] Announcer: This law requires bots to regularly disclose their non-human nature to users. [07:57] Announcer: It also mandates that operators make reasonable efforts to detect suicidal ideation [08:02] Announcer: and self-harm among their users. It was an early recognition that the drift of a system can have [08:08] Announcer: profound emotional and psychological impacts on the people who interact with it. [08:14] Announcer: Earlier this October, Governor Hochul signed yet another layer of protection, [08:19] Announcer: specifically for minors. This law, which will take effect on June 1st, 2027, [08:25] Announcer: restricts behaviors like deceptive human-like interactions and emotionally manipulative [08:31] Announcer: or sycophantic responses. It specifically bans the encouragement of behaviors like disordered eating [08:37] Announcer: or self-harm. The state is identifying the specific symptoms of a system that has drifted away [08:43] Announcer: from being a tool and into becoming an influencer. These specific interventions highlight [08:49] Announcer: where the drift has already been observed. The state is not creating these laws in a vacuum; [08:55] Announcer: they are responding to behaviors that have already been documented in deployed systems. [09:00] Announcer: The ban on sycophantic responses, for instance, acknowledges a known technical tendency [09:05] Announcer: of large language models to mirror and reinforce an agent's biases or emotional states, even [09:12] Announcer: when those states are harmful. [09:14] Announcer: The law is trying to build a barrier where the technology has failed to provide its own. [09:19] Margaret Ellis: By the time the June 2027 law takes effect, these behaviors will have been part [09:25] Margaret Ellis: of the operational reality of these models for years. This is the core of operational [09:31] Announcer: drift: the behavior is normalized through use, and the regulation arrives only [09:36] Announcer: after the social cost has become undeniable. The state is now trying to move from reactive laws [09:43] Announcer: for miners to a proactive framework for the frontier models themselves. [09:48] Announcer: They are attempting to move upstream of the problem before the harm becomes systemic. [09:53] Announcer: But the transition to proactive oversight relies on the newly established Digit Office [09:59] Announcer: having the technical capacity to actually understand what they are reviewing. [10:04] Announcer: The $55 million allocated to the Binghamton University Center is intended to provide that capacity. [10:10] Announcer: It is an investment in the state's ability to look under the hood of systems [10:15] Announcer: that companies often claim are too complex for outside audit. The question is whether $55 million [10:22] Announcer: is enough to match the research budgets of the trillion-dollar entities they are monitoring. [10:27] Announcer: We must look at the incentives involved. For a company with over 500 million dollars in revenue, [10:33] Announcer: a three million dollar fine is a marginal expense. The real deterrent is the transparency report [10:39] Announcer: itself. If these reports reveal a pattern of critical safety incidents [10:43] Announcer: that the companies failed to mitigate, the liability shift could be significant. [10:48] Announcer: When the failure is on the record, the excuse of unforeseen drift no longer holds. [10:54] Announcer: The report becomes the evidence that moves responsibility from the machine to the builder. [10:59] Announcer: This is why the registration portal launch on November 10th was more [11:03] Announcer: than just a bureaucratic milestone. It was the moment the state demanded a ledger. [11:08] Announcer: For the first time, OpenAI and Anthropic must commit to a public record of their safety operations [11:14] Announcer: in New York, or face the unspecified enforcement actions. The state spokesperson declined to detail. [11:20] Announcer: The ambiguity of that threat may be more powerful than a specific fine ever could be. [11:26] Announcer: It leaves the labs wondering exactly how much the state is willing to intervene. [11:30] Announcer: The drift here is not just in the technology, but in the governance. We are seeing a shift [11:37] Announcer: from federal inaction to aggressive state-level experimentation. California, Illinois, [11:44] Announcer: and New York are each pulling the frontier developers in slightly different directions, [11:49] Announcer: creating a regulatory environment where the most stringent state becomes [11:54] Announcer: the de facto national standard for reporting speed. If you have to report in 72 hours in New York, [12:01] Announcer: you are effectively reporting at that speed for the entire nation. In the press briefing, [12:07] Announcer: Governor Hochul remarked that the vision of AI as inevitable is something the state is no longer [12:13] Announcer: willing to accept without question. The creation of the Office of Digital Innovation, Governance, [12:20] Announcer: Integrity and Trust is an institutionalization of that skepticism. It places a permanent monitor [12:26] Announcer: on the institutions that are building what they hope will be the next foundation [12:30] Announcer: of the global economy. It is a statement that the future is not a foregone conclusion, [12:35] Announcer: but a space that must be governed. The question that remains is how the systems will adapt to [12:41] Announcer: being watched. If a developer knows they have only 72 hours to report a critical incident, [12:47] Announcer: the incentive may shift toward narrowing the internal definition of what is considered critical. [12:53] Announcer: The drift doesn't stop because a law is passed; it simply changes its path to avoid [12:58] Announcer: the new obstacles. We must watch the definitions as closely as we watch the systems themselves, [13:04] Announcer: or we risk a transparency that is technically accurate but functionally useless. [13:10] Announcer: As we approach the January 1st deadline, the focus will be on who registers [13:15] Announcer: and what their initial safety frameworks reveal. The Raise ISE Act assumes [13:20] Announcer: that transparency leads to safety, but transparency is only as good as the metrics being measured. [13:26] Announcer: If the state of New York and the researchers at Binghamton cannot agree with the labs [13:30] Announcer: on what constitutes a safety failure, the portal will only record the drift, not stop it. [13:36] Announcer: A record of a disaster is not the same thing as a disaster prevented. The record shows [13:42] Announcer: that the state is prepared to act against those who fail to register. This implies that the era [13:47] Announcer: of ask for forgiveness, not permission, is being replaced by a formal registration of intent. [13:53] Announcer: But intent and outcome are rarely the same in complex autonomous systems. [13:58] Announcer: The portal is an attempt to hold a company to its intentions, even when the code begins to pursue [14:02] Announcer: its own logic. It is a legal anchor for a technical ship that is already beginning to sail away. [14:08] Announcer: We are currently observing the beginning of a long-term tension between the velocity [14:14] Announcer: of software development and the friction of digital governance. The 72-hour reporting requirement is [14:21] Announcer: the shortest window yet imposed on these companies. It is a direct challenge to the idea [14:26] Announcer: that these systems are too fast or too complex to be held to a human timeline. [14:31] Announcer: It asserts that if a system operates in the world, it must operate within the world's ability [14:37] Announcer: to react to its failures. The $55 million at Binghamton, the new digit office, [14:44] Announcer: and the revenue-based thresholds are the infrastructure of a new kind of oversight. [14:49] Announcer: But as the miner safety law of June 2027 reminds us, the harms are often already baked into [14:55] Announcer: the system by the time the restrictions are signed into law. Deceptive human-like interactions [15:01] Announcer: were already a feature before they were a violation. We are consistently regulating [15:06] Announcer: the rearview mirror while the car continues to accelerate toward the next unknown. [15:12] Announcer: Operational drift isn't the point where something breaks; it's the point where the break [15:17] Announcer: is accepted as normal operation. By standardizing the reporting of these breaks, [15:22] Announcer: New York is trying to ensure they are never viewed as normal. They are trying to ensure [15:26] Announcer: that the AI cliff is visible before the descent begins. It is an effort to keep the stakes high, [15:33] Announcer: to prevent the slow erosion of safety from becoming a quiet catastrophe that we simply learn to live [15:39] Announcer: with. According to the TechCrunch reporting, the registration portal is now live. The countdown to [15:45] Announcer: January 1st, 2027 has begun. The labs have been notified. The financial thresholds have been set. [15:53] Announcer: The Office of Digit is waiting for the filings. It is a period of quiet anticipation, [15:58] Announcer: where the bureaucratic gears are turning, while the industry decides how much it is [16:02] Announcer: willing to disclose. The silence from the frontier developers is at this moment [16:07] Announcer: the most significant part of the record. We must consider what happens if the transparency reports [16:13] Announcer: are empty. If a company like OpenAI or Anthropic reports zero critical safety incidents over [16:19] Announcer: a full year, does that indicate a perfectly safe system, or a reporting framework [16:24] Announcer: that has drifted so far from reality that it is no longer functional? An empty ledger can be as [16:30] Announcer: telling as a full one. It can signal a failure of the metrics themselves, [16:34] Announcer: or a deliberate effort to keep the most significant drifts off the public record entirely. [16:40] Margaret Ellis: The state has claimed it will evaluate these frameworks independently through the Binghamton Center. [16:47] Margaret Ellis: This sets up a potential conflict of evidence: the company's internal safety report [16:52] Margaret Ellis: versus the state's external evaluation. [16:55] Margaret Ellis: In that conflict, the definition of responsibility will be tested. [17:00] Announcer: Who do we believe [17:01] Margaret Ellis: when the builder says the system is stable, [17:04] Announcer: but the auditor [17:05] Margaret Ellis: says it [17:05] Announcer: is drifting? This struggle over the truth of the system is where the real governance [17:11] Announcer: of artificial intelligence will occur. As these state-level laws proliferate, the divergence in [17:17] Announcer: standards—15 days in California versus 72 hours in New York—creates a regulatory noise [17:24] Announcer: that may allow smaller drifts to go unnoticed. The labs may find it easier to comply with the letter [17:31] Announcer: of the most frequent reporting standard while losing the spirit of safety [17:35] Announcer: in the administrative overhead. We risk creating a system of compliance that is [17:39] Announcer: so busy checking boxes that it fails to notice when the machine is no longer listening. [17:46] Announcer: The RAISE Act is a model, but it is also a test. It is a test of whether a state government [17:53] Announcer: can effectively oversee the most complex technology ever developed, or whether the technology [17:58] Announcer: will simply drift into new forms that the law was not written to cover. [18:03] Announcer: It is a contest between the permanence of law and the fluidity of software. [18:08] Announcer: If the law cannot adapt as fast as the drift, then the portal will become a monument [18:13] Announcer: to a regulatory intent that was left behind. We are left with a fundamental uncertainty. [18:19] Announcer: As companies register their frontier systems in New York, are they submitting to oversight, [18:24] Announcer: or are they simply formalizing a process where they continue to set the terms [18:28] Announcer: of their own transparency? The act of registration creates the appearance of control, [18:34] Announcer: but real control requires a level of technical depth that the state is still trying to build. [18:40] Announcer: We are witnessing the construction of the cage, but the lion is already much larger than the bars. [18:46] Announcer: Responsibility doesn't disappear; it relocates. In this case, it is relocating from the private labs [18:52] Announcer: to a state registration portal. But the people behind the portal still depend on the labs [18:57] Announcer: to tell them when something has gone wrong. This dependency is the weak link in the chain [19:02] Announcer: of accountability. If the source of information is also the source of the risk, [19:07] Announcer: the record will always be incomplete. [19:10] Announcer: We are relying on the industry to report its own descent, which is a fragile basis for safety. [19:16] Margaret Ellis: We still do not know how DIGIT will define a violation worthy of a three million dollar fine, [19:24] Announcer: or what [19:25] Margaret Ellis: specifics were kept [19:26] Announcer: in the dark by the state spokesperson regarding enforcement for those who do not register. [19:32] Margaret Ellis: The ledger is open, [19:34] Announcer: but the most critical [19:35] Margaret Ellis: entries have yet to be made. [19:37] Announcer: We are waiting for the first admission [19:39] Margaret Ellis: of failure—the [19:41] Announcer: first moment the 72-hour clock starts ticking—and we see if the system is capable of telling us [19:47] Announcer: the truth about its own drift. For more information on the documents and findings in this case, [19:53] Announcer: visit operationaldrift.neuralnewscast.com. Neural Newscast is AI-assisted, human-reviewed. [20:02] Announcer: View our AI transparency policy at neuralnewscast.com. This record is closed. Neural Newscast [20:10] Announcer: is AI-assisted, human-reviewed. View our AI transparency policy at neuralnewscast.com.