Yuval Avidani
Author
On October 5, 2026, representatives from OpenAI, Anthropic, Google and Meta sat down in front of the New York City Council, swore to tell the truth, and listened to ten proposed bills written specifically about them. The fifth company summoned, SpaceX AI, didn't show up despite an official summons. That same quarter, startups in New York raised $8.88 billion, and half the money went to AI. New York is right now both one of the hottest markets in the world for AI builders, and one of the places writing laws about them the fastest.
I'm building a presence in New York these days, so I sat down to put together one map for myself (and for you): what the law actually says, who it really applies to, and where it's worth showing up next quarter. Important disclaimer up front: this is not legal advice. This is a builder's map for builders, and any real step should be taken with a lawyer who knows this space.
The example we'll carry through: HireLoop from Brooklyn
So this doesn't stay theoretical, let's invent a startup. We'll call it HireLoop (a completely made-up name). Twelve people in Brooklyn building a hiring assistant: a system that reads resumes, ranks candidates, and runs an initial screening chat conversation with them. Under the hood, they haven't trained any model themselves — they're calling the API of a frontier model (one of the strongest models on the market, like Claude or GPT). Their customers are employers in New York, and lately they're also considering a follow-up product: a learning assistant for schools.
Think of HireLoop like a restaurant buying raw ingredients from a big supplier. There are laws about the supplier (how things are grown and shipped), laws about the restaurant (hygiene, allergen labeling), and laws about who's allowed to serve what to whom. In New York there are exactly three such layers: a state law about the big suppliers, city laws about anyone using AI to make decisions about people, and rules from the school system. Let's go through them one by one, and each time see what it means for HireLoop.
Layer 1: The RAISE Act, the law for the big suppliers
The RAISE Act (short for Responsible AI Safety and Education) is a New York State law that puts transparency and safety obligations on the companies training the biggest models in the world, on behalf of the public that lives with the results. The law was first signed on December 19, 2025, a final amendment was signed on March 27, 2026, and it takes effect on January 1, 2027.
Who clears the bar? A "frontier developer" is anyone who trained a model using more than 10^26 compute operations. To give you a sense of scale: that's a number with 26 zeros, an order of magnitude only a handful of labs in the world reach. A "large" frontier developer is one that also brings in over $500 million in annual revenue. The heaviest obligations fall on the big players:
- A public safety framework (frontier AI framework): a document explaining how the company identifies catastrophic risks, when it stops, and how it protects the model's weights. Updated at least once a year.
- Reporting a critical safety incident within 72 hours of detection, and if there's an immediate danger to human life, 24 hours to law enforcement.
- A transparency report before every launch of a new model, plus quarterly reporting on risk assessments from internal use.
- Registration and ownership disclosure with a new office within the state's Department of Financial Services (DFS), called DIGIT.
Penalties: up to $1 million for a first violation and up to $3 million for additional ones. On September 21, Governor Hochul and DFS announced that starting November 2026, large frontier developers will be required to register, with full compliance kicking in come January. And what about HireLoop? Based on the legal analyses, anyone building on an API without training a model at that scale doesn't get direct reporting obligations from this law. But that doesn't mean you can ignore it: the safety frameworks and transparency reports of our suppliers become public documents, and that's exactly the material an enterprise customer will ask about in a procurement questionnaire. Also — if HireLoop ever trains its own model on top of an open model, it's worth knowing that the compute used in that additional training counts toward the threshold.
Layer 2: The city, and where HireLoop actually lives
This is where the story gets personal. Local Law 144 is New York City's law on automated hiring tools (AEDT — software that scores or filters candidates), and it's been enforced since July 2023. The law requires an independent bias audit every year — meaning an external auditor calculates whether the system selects different gender and race groups at different rates — publication of a summary of the results, and notice to candidates at least 10 business days before use. The penalty: up to $500 for a first violation, and up to $1,500 for every additional one, with each candidate who didn't get notified counting as a separate violation.
Turns out enforcement so far has been weak. A December 2025 audit by the New York State Comptroller found that the city's enforcement was "not effective": the city department checked 32 companies and found one issue; the Comptroller's own team checked the same 32 and found at least 17 possible issues. Out of 12 test calls to the 311 hotline about the law, only 3 were routed correctly. The city agreed to implement most of the recommendations, and law firms like DLA Piper are already writing that the risk for employers is rising. For HireLoop, this is the most important line in the whole article: its customers are the ones required to run the audit, and they're going to come asking HireLoop for the data that makes it possible.
And what happened at the October 5 hearing? The Council's full committee, chaired by Council Speaker Julie Menin and Carmen De La Rosa, discussed ten proposals. Among the notable ones: third-party verification before marketing a model and a human "kill switch" capability, reporting AI safety incidents in city contracts within 24 hours, a private right of action for foreseeable harm from model misuse, restrictions on chatbots (notifying users, explicit consent before training on adult conversations, and a ban on training on minors' conversations), whistleblower protections, and a complaint mechanism with a reward for the reporting citizen. These are proposals, not laws, and none of them has passed. But if HireLoop is running screening conversations in chat, the chatbot proposal touches it directly, and it's worth tracking.
Another body worth knowing: the Office of Algorithmic Accountability, established by 2025 city legislation and supposed to be up and running by June 2026, to review algorithmic tools used by city agencies before they go live. According to a follow-up audit by the State Comptroller from August, the office is still not fully operational, with six budgeted positions. Anyone selling to the city itself will run into it soon.
Layer 3: Schools, the product that almost got built
Remember HireLoop's follow-up product? On September 2, 2026, Mayor Zohran Mamdani and Chancellor Kamar Samuels announced a one-year moratorium on generative AI aimed at students, from kindergarten through 8th grade. We're talking about nearly 600,000 students, two-thirds of the system, and according to the city this is the broadest policy of its kind in the US. "Companion"-style chatbots are banned at all ages. Teachers can still use AI for lesson planning, with tools that meet the system's safety standards.
High schools are a different story: every student will go through two AI literacy modules a year, 45 minutes each, covering what AI is, bias, risks, ethics, and career impact. On top of that, up to 50,000 high school students will use approved tools, in up to five classes per school. For HireLoop, the takeaway is simple: a product for New York elementary school students is off the table for the coming year, but a tool for teachers or for high school that passes a safety review is fair game.
The checklist: what HireLoop does next quarter
Now let's turn all this into actual work. This is a list any team building an AI product in New York can copy and adapt, lawyer at their side:
In short, for those reading without the widgets: map out which model we're consuming from and pull the supplier's safety framework, prepare the data customers will need for their bias audits ahead of time, write a candidate-notice template, add notice and explicit consent in chat and stop training on conversations without consent, build an internal incident procedure, filter out elementary-school customers, and track the Council's proposed bills.
Where to show up: the community and the money
The laws are one side of the story. The other side is that New York is simply boiling right now. According to AlleyWatch, in Q2 2026 startups in the city raised $8.88 billion across 233 deals, up 46% year-over-year — the strongest second quarter in the city's history. AI companies took $4.56 billion of that, 51%, across 81 deals. Twenty-four rounds of $100 million or more took 63% of the money, and frontier research labs like Flourish and General Intuition raised hundred-million-dollar-plus Series A rounds there.
So where do you actually meet all these people? Here's the lineup, checked against the official sites:
- AI Tinkerers NYC: a builder community where people show real code and real systems, not sales decks. October 21 is a Demo Day on agent loops, October 24 is a walking hackathon with Twilio, and November 7 is a fashion-and-fintech hackathon.
- AI DevCon NYC, November 2–4, Industry City in Brooklyn: Tessl's conference on building with code agents at team scale.
- AI Dev NYC, November 30 – December 1, Chelsea Piers: DeepLearning.AI and Andrew Ng's conference, roughly 2,500 engineers and builders.
- The AI Summit New York, December 9–10, Javits Center: the enterprise conference, over 7,500 attendees.
- QCon AI New York, December 15–16, Jersey City (one PATH stop from Manhattan): AI in production, token costs, evals, and permissions for agents.
If HireLoop asked me, I'd send the engineers to AI Tinkerers and AI Dev, and the CEO to AI Summit, where the enterprise customers are sitting.
My take: the direction is clear, the details are still moving
In my view, New York is doing something interesting: the state is regulating the handful of big labs, and the city is regulating whoever uses AI to make decisions about people — about employees and about kids. For a startup like HireLoop, that's actually good news. The heavy obligations fall on the suppliers, and our obligations are mostly about transparency toward the customer and the user. A team that builds this in from day one — with documentation, notices, and audit-ready data — will find it's a selling point, not a tax.
It's also important to state the limits here. The Council's ten proposals could change dramatically or not pass at all. The practical details of registering with DIGIT and of the assessment fees haven't been fully published yet. And the Office of Algorithmic Accountability is still being stood up. If you're building here, you need to keep watching, not guessing.
And to close, on a personal note: I'm building a presence in New York right now, and I'd genuinely love to meet builders, startups and companies in the city. If you're there, message me on LinkedIn, and maybe we'll run into each other at one of the conferences on this list.
So if HireLoop were our product, which of these steps would we actually do this week?
Want to go deeper with this?
If you want to really understand how frontier models work, and why we even measure them in compute operations, you'll find it in the AI Fundamentals course, in English.
If you're building products and automations with Claude, including MCP, Skills and agents, you'll find all the hands-on work in the flagship course in Hebrew, Practical AI with Claude.
I post updates from the ground in New York, and demos, on YouTube and LinkedIn, and in short form on X, Instagram and TikTok. My code is open on GitHub.