NY Hearing: Anthropic, OpenAI, Google, Meta Testify

Representatives from Anthropic, OpenAI, Google, and Meta testified under oath on Monday (5) before the 51 city council members of New York regarding the risks of artificial intelligence. The meeting of the City Council's Full Committee served as a hearing for a package of bills that would require external validation of AI systems prior to commercial use in the city.

Anthropic sent Logan Graham, head of the Frontier Red Team. OpenAI was represented by Morgan Dwyer, responsible for public policy development and operations; Google by Alice Friend, director of AI policy and emerging technologies; and Meta by Shane Cahill, director of AI policy for legislation. Anthropic, OpenAI, and Google agreed to attend only after the Council threatened subpoenas, according to coverage from CNBC and Bloomberg. Meta had confirmed its participation beforehand.

Who Didn’t Show and Who Was Vocal

SpaceXAI, Elon Musk's AI branch, was subpoenaed but did not appear. Council President Julie Menin labeled the absence as a "direct violation" of the subpoena and stated that the case will go to court; the Council indicated it may seek enforcement in the New York State Supreme Court.

The harshest tone came from former employees. Jacob Coxon, an engineer who left Anthropic and voiced his concerns weeks ago, described the industry's stance as "extremely reckless" and stated he considers it "more likely than not" that humanity will lose control of these systems. Alex Turner, a former researcher at Google DeepMind, advocated for greater transparency and a slower pace of development. Company executives, according to the Associated Press, claimed that their systems are already enhancing people's daily lives and that safety is a priority. AMNY reported that all four companies avoided providing clear answers to several of the Council members' key questions.

What’s Up for Vote

The package presented by Menin on September 25 has four main components. The broadest prohibits the marketing or deployment of an AI system in the city without prior validation from an independent reviewer, covering data quality, bias, privacy, security, and model outputs, and also requires a human shutdown mechanism, the so-called kill switch. Another bill mandates that contractors and municipal agencies report AI security incidents to the city’s Cyber Command within 24 hours. A third allows whistleblowers to receive a portion of fines levied against AI companies. The fourth enables lawsuits against developers for damages caused by third parties who bypass security controls, provided that the damage was foreseeable and reasonable safeguards were lacking.

Why This Matters to CIOs, Not Just Legal

New York is home to JPMorgan, Citi, Goldman Sachs, and Morgan Stanley, as well as offices of the Big Four audit firms. If the validation bill passes in its current form, a bank deploying an AI agent to assist customers in the city would need to obtain an external review before activating the system. This creates an immediate market for AI auditing and a queue for approvals that currently doesn’t exist in any CIO’s budget.

The 24-hour deadline for incident reporting also stands out in international comparison. California’s SB 53 gives frontier model developers up to 15 days to report a critical security incident, reduced to 24 hours only when there is imminent risk of death or serious injury. In the European Union, Article 73 of the AI Act sets a 15-day timeline for serious incidents and up to two days for those affecting critical infrastructure. Suppliers servicing New York City, including consultancies and integrators, would face the shortest timeline of the three regions.

The Counterargument

The most consistent criticism of the package focuses on competence and scale. A municipal rule concerning a globally distributed product tends to create fragmented requirements, and Menin herself questioned during the hearing how to balance the development of frontier models with the race against China. Advocates respond by highlighting the federal void: AP notes that Congress has not enacted AI security legislation, and the Trump administration took a passive stance, pushing states and cities to the forefront.

There is also a practical limitation. A City Council has subpoena power, as was made clear on Monday, but cannot inspect the training of a model located in another state. The validation outlined in the bill targets use within the city, not development.

For companies operating in New York, California, and Europe, the likely outcome is a compliance map with three timelines and three different definitions of incident for the same system. Monday’s hearing did not decide anything, but it demonstrated that the city is willing to use subpoenas against AI labs, all while being home to the headquarters of the largest banks in the country.