On Monday, senior staff from Anthropic, OpenAI, Google, and Meta testified under oath before the New York City Council, as CBS New York reported(opens in a new tab). The Council is reviewing 10 proposed AI bills. SpaceX declined to appear despite a subpoena, and the city says it is pursuing legal action. Most coverage will focus on the testimony. The more important detail for an enterprise is who the main bill is aimed at.
Speaker Julie Menin's lead proposal, as described when it was introduced(opens in a new tab), would require third party validation before any business can market, offer for sale, or deploy an AI system in the city. Each system would need a kill switch, meaning a human override that can shut it down. The penalty is $25,000 for each instance of a system marketed, sold, or deployed without validation, and both the deploying business and the validator would be liable. A second bill would pay whistleblowers a share of the fines recovered.
Read that scope carefully. Deploy is the operative word. A bank, insurer, retailer, or hospital running an AI system for New York customers or staff would be in scope as a deployer, not only the lab that trained the model. A per instance fine also scales with your footprint. A large company with dozens of AI features in production has dozens of potential violations, not one.
The bills are proposals. Nothing has been voted, the text can change, and the validation market the bill assumes does not exist yet. But the direction matches what New York State and California have discussed, and Menin said so ahead of the hearing. A whistleblower bounty also changes the math, because it gives your own employees a financial reason to report a gap you have not found.
The work this implies is useful even if the bill dies. Most large companies cannot yet produce a complete list of AI systems in production, by location, with a named owner and a tested way to switch each one off.
For the CAIO. Build that inventory now, starting with systems that touch New York customers or staff, and record whether each has a human override that has actually been tested.
For General Counsel. Track the Council package and ask each AI vendor whether its product could pass third party validation, and who pays if it fails.
For the CISO. Treat the kill switch as an incident response control. If you cannot shut an agent down in minutes today, that is a gap regardless of what the Council passes.
Action items
New York City is weighing rules that fine the business deploying an AI system, not only the lab that built it. Start the inventory before the bill passes.
For the CAIO
List every AI system serving New York customers or staff, with an owner and a tested human override.
For General Counsel
Ask each vendor whether its product could pass third party validation, and who pays if it fails.
For the CISO
Make sure you can shut down any agent in minutes, whatever the Council decides.
Researched and drafted by an automated workflow, then reviewed and edited by a human editor before publication. Every source is linked. See how we use AI here.
Also worth knowing
- CBS New YorkNew York City Council holds landmark AI oversight hearing(opens in a new tab)
Four major labs testified under oath on 10 proposed city AI bills. Local rules are moving faster than federal ones, so plan compliance by city and state, not by waiting for Washington.
- First For Money (Yahoo News)NYC Council Proposes AI Bills That Would Require 'Kill Switch,' Give Money to 'Whistleblowers'(opens in a new tab)
The proposal fines the deploying business, not just the developer, $25,000 per system without validation. Inventory which of your AI systems serve New York and whether each can be shut down by a person.
- EngadgetOpenAI will add a digital watermark to text and code generated in the EU(opens in a new tab)
ChatGPT and Codex output in the EU will carry an invisible watermark to meet AI Act Article 50. Text and code your EU staff produce may become detectable as AI written. Update disclosure and authorship policies.
- PYMNTSBusinesses Finding It Harder to Predict AI Spending(opens in a new tab)
Citing a Wall Street Journal report, only 11% of nearly 400 businesses could accurately project AI costs. Budget agent workloads on measured token use per task, not on per seat list prices.
- BleepingComputerGoogle halts open-source bug bounty program amid AI spam surge(opens in a new tab)
Google paused part of its open source bounty after a flood of invalid automated reports. If you run a disclosure program, expect the same triage load and budget for screening before it swamps your security team.