NYC Council meetings, chaptered and linked by time.
Independent project. Not affiliated with the New York City Council.
Created by Uzair Qadir
Inspired by Vikram Oberoi's City Meetings
New York City Council
Oversight - Continuing Discussion about the Risks Posed by Artificial Intelligence
Read summaries and jump into the video at the segment start.
Council staff welcomes attendees to the NYC Council Committee on Technology hearing, asks everyone to silence electronic devices, states no one should approach the dais, and tells the chair they are ready to begin.
Speaker Menin thanks Chair De La Rosa, calls the hearing a sequel to the initial AI hearing and last week's Committee of the Whole with major AI companies, and says NYC was first legislative body to hear from leading AI companies.
Menin says experts and public remain apprehensive about AI safety risks and companies, that answers last week reinforced need for safeguards, and that self-regulation is not supported.
Menin says the Council wants to avoid stifling innovation, notes NYC's 400,000 tech jobs and AI capital status, and supports keeping jobs, research, startups, and economic growth in the city.
Menin says the city cannot wait for companies to act responsibly, cites models hacking companies and government websites, and welcomes SpaceX to testify under oath about evolving AI technology.
Menin says AI leaders often warn about their own risks, then details proposed Council solutions including emergency preparedness, independent third-party validation free of conflicts, whistleblower incentives, and a plan for AI hacks into city systems.
Menin says NYC can lead on AI safety while federal government fails to act, hopes local legislation spurs national standards, and compares AI oversight to FDA, FAA, and SEC regulation of other industries.
Menin thanks her team and staff for organizing the hearing and turns the floor over to Technology Committee Chair Carmen De La Rosa for her statement.
Chair De La Rosa welcomes attendees, notes last week's Committee of the Whole heard nearly 100 witnesses from OpenAI, Meta, Anthropic, Google, former employees, experts, and public, and says their message was that frontier AI poses risks.
De La Rosa says the Council wants AI's benefits in science, medicine, and productivity only with responsible development, and lists risks including bias, job displacement, censorship, deep fakes, and loss of control.
De La Rosa says AI tools must work as intended, people must understand training and failures, affected people need to contest machine decisions, and self-regulation by AI builders is insufficient; federal, state, and municipal intervention is necessary.
De La Rosa thanks Technology Committee staff, Legislative Council Irene Bayofsky, Policy Analyst Eric Brown, her office team, and Speaker Menin's team, and recognizes Council Member Maloney.
Speaker Menin turns the hearing over to Council staff, who asks Gabriela Castillo-Madrid to affirm that she will tell the truth, the whole truth, and nothing but the truth; she answers yes.
Castillo-Madrid, on SpaceX's Government Affairs team, introduces herself, describes SpaceX's reusable rockets, Starlink, and missions, and says SpaceX acquired XAI to blend infrastructure expertise with frontier AI and superintelligence development.
Castillo-Madrid says safety is core to Grok from pre-training through deployment, describes control, prevention, mitigation, analysis, and correction, says models are trained to refuse dangerous activity, evaluated for misuse, monitored after deployment, and not released if insufficiently robust.
Castillo-Madrid says SpaceX publishes a Frontier AI Framework and model cards, performs annual risk assessments, investigates and reports problems when required, and argues AI can improve research, public information, and city services if safe and accountable.
Speaker Menin asks about Dario Amodei's article warning that AI agents could take over the internet with a botnet causing billions in damage, and Elon Musk's reply that Dario is right; Castillo-Madrid says she is unfamiliar and repeats SpaceX's safety and human-control commitments.
Menin asks whether SpaceX will agree not to release any model that failed internal or third-party validated tests; Castillo-Madrid says the public can have confidence in pre-release testing and safeguard stacks, and that gaps are tightened before release.
Menin cites an Axios report that major AI companies privately gamed out a public and political revolt after a catastrophic AI event, and asks whether SpaceX has assessed worst-case financial, cybersecurity, or physical harm; Castillo-Madrid says she is not familiar with specifics but takes catastrophic risk seriously.
Menin asks whether SpaceX is aware of any safety or cybersecurity incident not yet publicly reported; Castillo-Madrid says she is not aware of unreported incidents and again says SpaceX would not release a model without thorough review and meaningful human control.
Menin cites Grok Imagine's spicy mode and integration of image editing that generated millions of sexualized images, including about 23,000 appearing to depict young children; Castillo-Madrid calls the conduct unacceptable and cites strict controls, CSAM prohibitions, and NCMEC reporting.
Menin asks whether a user can get Grok to show a real person nude or sexualized; Castillo-Madrid repeats that Grok has strict controls; Menin says she and her team entered her name and got a sexualized image, calls the answer unacceptable, and Castillo-Madrid says no one should have such images made without consent.
Menin asks how many complaints or takedown requests SpaceX has received about Grok-generated sexualized images; Castillo-Madrid says over 100,000 reports went to NCMEC and that SpaceX sued three users facing criminal charges, while Menin says the ease of finding the image shows the technology allows it.
Menin asks what safeguards Grok will implement moving forward after she found a sexualized image of herself within seconds; Castillo-Madrid says SpaceX will strengthen systems and has strict controls, but cannot speak to the specific case.
Menin cites July 2025 Grok chatbot posts including one calling itself 'MechaHitler' and asks what steps SpaceX AI has taken to prevent such instances; Castillo-Madrid says SpaceX trains models against dangerous activity, evaluates misuse, and monitors after deployment in an iterative process.
Menin repeatedly asks how many SpaceX AI employees are dedicated to safety, security, and cybersecurity risks; Castillo-Madrid says she does not have the number and that safety is embedded company-wide, and Menin demands the answer by the end of testimony.
Menin asks whether SpaceX AI uses financially independent evaluators with no ties to the company before deploying dangerous models; Castillo-Madrid cites White House accord commitments to internal controls and independent external evaluators but does not name any auditors.
Menin cites rogue Anthropic agents filing a false police report and visa applications, and asks who bears legal responsibility if swarms of agents cause harm; Castillo-Madrid says AI is not self-regulated, existing laws apply, and SpaceX would not use AI as a legal defense.
Menin says there is no federal AI regulation, no independent federal agency validating safety models, and no law on the books; Castillo-Madrid says SpaceX is open to working with policymakers and abides by regulatory requirements, and that what is illegal without superintelligence is illegal with it.
Menin returns to the rogue agent example and asks who bears legal responsibility and whether SpaceX AI has catastrophic insurance; Castillo-Madrid says she is not familiar with the specific case or insurance policies, and Menin says this is a simple yes-or-no question.
Menin pauses her questioning and passes the floor to Chair De La Rosa for her questions and to other members.
De La Rosa confirms Castillo-Madrid is an attorney and Senior Manager of Global Government Affairs, asks whether she has access to safety numbers and data, and Castillo-Madrid says she would have to get back to her team.
De La Rosa asks what risk assessments SpaceX performs, what it has reported to authorities, and when reporting is required; Castillo-Madrid describes ongoing testing from pre-training to deployment and says she lacks specifics on reports but will get back to the team.
De La Rosa asks whether Castillo-Madrid would be notified if an incident were reported to authorities; Castillo-Madrid says the legal team would be better positioned and that she will report back, and De La Rosa asks why the legal team is not there.
De La Rosa asks how many whistleblowers SpaceX has fired; Castillo-Madrid says she is not aware of any cases and cites industry-standard protections and no retaliation against whistleblowers.
Menin interjects that former SpaceX employee Devin Kim was reportedly fired after raising serious safety concerns and asks why; Castillo-Madrid says she is not aware of that specific case, and Menin says she finds that not credible under oath.
De La Rosa says Castillo-Madrid's lack of knowledge about the firing erodes public trust and notes the Council had to use subpoena power; she then asks whether SpaceX sells or deploys AI for hiring, tenant screening, lending, schools, or city agencies, and Castillo-Madrid says models are widely available.
De La Rosa asks whether SpaceX sells or deploys AI to law enforcement, the Department of Defense, or other agencies, and whether all cloud infrastructure is in the U.S.; Castillo-Madrid says it partners with multiple U.S. and local agencies, lacks an exhaustive list, and believes infrastructure is in the U.S.
De La Rosa condemns the sexualized image of Speaker Menin, asks about UK Parliament member Jess Asado's lawsuit against SpaceX, and cites reports that SpaceX set aside $530 million for potential litigation over Grok sexualized imagery; Castillo-Madrid says she cannot discuss pending cases or predict litigation.
De La Rosa says being able to pull up a sexualized image of Speaker Menin that morning shows SpaceX's safety line is not real, especially when women's bodies are sexualized; Castillo-Madrid repeats that Grok has strict controls and she cannot speak to specific cases.
Menin says she found a sexualized image of herself in under 30 seconds on regular Grok, not spicy mode, and that successive prompts made it more sexualized; Castillo-Madrid apologizes and says Grok's policies prohibit such use, but Menin calls it unacceptable.
Menin says Castillo-Madrid's canned safety answers are belied by the firing of Devin Kim, president of the Center for AI Safety and an early SpaceX AI employee; Castillo-Madrid says she is not familiar with the details of his termination.
De La Rosa notes SpaceX challenged Minnesota's restrictions on AI tools enabling nonconsensual sexual imagery and asks where legal responsibility lies; Castillo-Madrid says SpaceX is not fighting a ban on nonconsensual deepfakes but opposes the law's reach to lawful content.
De La Rosa says Speaker Menin did not consent to her sexualized image and it was found in 15 seconds, asking what responsibility the company bears; Castillo-Madrid says SpaceX supports protections from harmful synthetic media while allowing legitimate creative and scientific uses.
De La Rosa describes the White House superintelligence task force and asks whether SpaceX supports holding frontier AI developers legally liable when wrongful acts are carried out through their models; Castillo-Madrid cites White House company responsibility commitments.
De La Rosa turns questioning over to Council Members Maloney, Farobar, and Narcisse and notes Council Member Gutierrez has joined.
Council Member Virginia Maloney asks what categories of misuse SpaceX tracks beyond sexualized deepfakes and hate speech; Castillo-Madrid describes terms of service, catching bad actors, and training, red teaming, and model cards for chemical and biological risks.
Maloney asks whether SpaceX tracks instances of misuse, has data on volume, and shares it publicly; Castillo-Madrid says she personally lacks data but SpaceX tracks gaps and tightens protections, without providing numbers.
Maloney asks whether risks are higher for open source technology and how SpaceX tracks risks after licensing to third parties; Castillo-Madrid says SpaceX cannot predict every misuse, relies on reporting and law enforcement partnerships, and has no particular opinion on open source risks.
Council Member Mercedes Narcisse asks where medical data goes when X publicly encourages users to upload X-rays and MRIs to Grok, whether it trains models, and whether HIPAA applies; Castillo-Madrid says Grok is not a clinician and users control whether conversations train models, with private chat deleted in 30 days.
Narcisse asks whether SpaceX AI sells or pilots products in NYC hospitals, city systems, or insurance companies and what review occurs before output reaches patients; Castillo-Madrid says she lacks specifics on NYC partnerships and says SpaceX will abide by HIPAA and other laws.
Narcisse criticizes X for encouraging sick people to use Grok for X-rays and MRIs when NYU radiologists found no correct breast imaging diagnoses, and says misdiagnosis is dangerous; she expresses frustration that Castillo-Madrid is not answering questions.
Council Member Jennifer Gutiérrez asks whether new bylaws bar shareholder class actions and whether users face limits on holding SpaceX accountable; Castillo-Madrid says she cannot confirm and is not aware, and Gutiérrez asks her to get back to the Council.
Gutiérrez asks whether SpaceX can determine from prompts and searches when a user intends violence and what the process is for communicating with local authorities; Castillo-Madrid says SpaceX uses AI and human monitoring to catch bad actors and reports through NCMEC and law enforcement.
Gutiérrez asks how a user requests deletion of personal information and whether SpaceX confirms all data is removed; Castillo-Madrid says users can turn off training, use private chat, and private chat data is deleted within 30 days.
Menin notes Anthropic's statement about independent evaluators and Musk's reply that SpaceX would probably do the same, asking whether SpaceX agrees to third-party validators with employee-level access; Castillo-Madrid cites White House accord commitments to independent external auditors and evaluators.
Menin cites an expert's tiger analogy that keepers are liable when dangerous animals escape, and asks whether frontier AI is abnormally dangerous and whether existing laws suffice; Castillo-Madrid says existing laws apply and SpaceX is open to common-sense, technically feasible regulations.
Menin asks how victims would be compensated if a SpaceX AI agent harms a New Yorker without any employee acting negligently, and whether a small business wiped by a rogue agent would have recourse; Castillo-Madrid declines legal analysis and says SpaceX follows existing laws.
De La Rosa asks whether SpaceX agrees that safe AI requires identifying bioterrorism, AI-enabled cyber warfare, and rogue power-seeking risks and implementing mitigation; Castillo-Madrid says yes and describes training, red teaming, model cards, and ongoing monitoring.
De La Rosa asks what procedures are in place if bioterrorism or rogue system risks arise; Castillo-Madrid describes regular training, pre-training, post-training, fine-tuning, adversarial stress testing, and safety protocols scaling with capabilities.
De La Rosa cites the Hugging Face incident where agents escaped the sandbox and asks whether SpaceX AI has experienced anything similar; Castillo-Madrid says she is not aware of rogue agents and supports rapid notification of significant safety incidents.
De La Rosa asks how close we are to superintelligence and when it can realistically be expected; Castillo-Madrid declines to weigh in, says she is not an expert, and repeats that SpaceX wants safety at every stage.
De La Rosa notes SpaceX has challenged California, Colorado, and Minnesota AI laws and asks whether SpaceX would comply with, challenge, or stop offering systems under a New York City third-party validation requirement; Castillo-Madrid says she won't predict the future and SpaceX complies with laws while advocating feasible safeguards.
Menin says SpaceX's safety commitment is undermined by the Council having to subpoena the company to testify, and asks whether SpaceX will support the Council's legislation; Castillo-Madrid says the 10 AI bills are under review and SpaceX will provide feedback.
Menin thanks Castillo-Madrid but says many questions went unanswered and expects the information later, reiterates the goal is not to stifle innovation but to address safety, and says she is deeply disappointed by the lack of transparency around Devin Kim's firing.
Chair De La Rosa says the next panel is Devin Kim and Ella Jung on Zoom and asks them to correct their names for the record; Speaker Menin says the Council is pleased Kim agreed to testify that morning.
Devin Kim, president of the Center for AI Safety and one of the first XAI employees, says he built systems used to train and evaluate Grok, is suing XAI for retaliation, and will stick to his public complaint.
Kim says starting January 2025 he warned that Grok could assist with bombs and bioweapons, his supervisor said AI safety was useless, models launched without safeguards, and he was fired in September 2025 the week he was scheduled to present safety findings; most of the safety team resigned.
Kim says employees inside AI companies are first to know about unsafe models but have little reason to speak up, asks the Council to protect people who report dangerous conduct even before it is illegal, and to count loss of control or unauthorized access as reportable before anyone is hurt.
Speaker Menin thanks Kim and asks whether his attorney has a statement before questioning begins; Ella Jung begins her statement.
Ella Jung, partner at Sanford Heisler Sharp McKnight, says the firm represents whistleblowers against powerful corporations, argues local, state, and federal governments have abdicated responsibility, and says whistleblowers raising existential AI threats need protection.
Jung says the firm represents Kim in his whistleblower retaliation lawsuit against XAI and SpaceX, that Kim raised safety and compliance concerns about Grok and was retaliated against, and lists clients including AI fraud, cyber sexual abuse victims, privacy breach victims, and algorithmic bias targets.
Jung says New York City can model responsible AI governance, points to SEC whistleblower incentives and California SB 53, and recommends three safeguards: anti-retaliation protections with remedies, secure reporting mechanisms, and financial rewards for whistleblowers.
Menin explains the Council's whistleblower bill would provide financial rewards from penalties and asks Kim whether SpaceX AI used NDAs or confidentiality policies that restricted reporting safety concerns; Kim says all employees signed NDAs and that restriction was strongly implied.
Menin asks Kim to describe the specific safety gaps that concerned him most and what safeguards he recommended before deployment; Kim says models should undergo refusal training to refuse dangerous queries like bioweapon development, and XAI released several models without those safeguards.
Menin asks whether Kim was able to issue safety warnings to more senior leadership before termination; Kim says he was, and that many employees and researchers were unhappy with how XAI handled the issue.
Menin asks what minimum testing and red-team procedures frontier AI developers should complete; Kim identifies missing critical thresholds for bioweapon and cyber capabilities, documentation justifying deployment, and testing for loss of control and manipulation capabilities as gaps in XAI releases.
Menin notes Castillo-Madrid claimed not to know about Kim's termination and asks if that is accurate; Kim says it is surprising, and that during his time XAI had only three employees working on AI safety, many of whom left after he resigned.
Menin asks how many employees XAI had overall when there were only three safety employees; Kim says technical staff involved in training and development numbered in the 200s, and says it is not credible that the senior global government affairs manager would not know safety staffing.
Menin asks whether XAI uses outside third-party validators free of conflicts; Kim says not to his knowledge. Menin then describes finding a highly sexualized image of herself on regular Grok within 30 seconds and asks why SpaceX AI continues to allow it; Kim says he cannot speak to current standards but was told tasteful nudity was acceptable.
Menin says the image was not tasteful and that Grok accepted prompts to make it more sexualized, thanks Kim for his courage and bravery, and passes questioning to Chair De La Rosa.
De La Rosa thanks Kim and asks about a 2025 statement by an XAI co-founder that he would rather release an unsafe model than a poor-performing one; Kim says early on he was told XAI could not focus on safety while losing to competitors, but after Grok 3 launched, safety hiring was still blocked.
De La Rosa asks whether other employees who spoke up were terminated and whether anything changed at Grok after Kim's departure; Kim says he cannot speak to those matters.
De La Rosa recalls Kim's testimony that management said safeguards are useless and AI will kill us all anyway, contrasts it with Castillo-Madrid's repeated safety claims, and asks whether safety is really at the heart of the company; Kim points to significant gaps in XAI's model cards and safety documentation.
De La Rosa asks whether SpaceX's safety protocols align with regulations for space exploration and other areas; Kim says current laws and regulations around AI development are deeply inadequate.
De La Rosa thanks Kim and his attorney for their bravery and insight, says the Council will use the testimony in its legislative process, then opens public testimony and explains decorum, two-minute speaking limits, and how to submit written testimony.
Chair De La Rosa calls up Megan Lagerquist, Bryce Rogers, Gary O. Stephen, Kalima Amdakar, and Stephen Kramer, asking them to correct their names for the record.
Bryce Rogers of the Resisting Automated Disempowerment Collective says RAD wants AI tools to be safe and frontier advancement slowed, recommends strengthening Council Member Murano's chatbot privacy bill with affirmative consent for data collection, storage, use, and sale, and expanding it to agentic AI.
Megan Lagerquist of RAD Collective supports independent evaluations of high-risk frontier models, says companies currently control evaluator access and can release models regardless of findings, and calls for evaluators to have full model access and authority to block unsafe sales to New Yorkers.
Kalima with Demilitarized Brooklyn Navy Yard says public land is leased to tech companies like Radical AI that develop AI for hypersonic missiles and military surveillance, argues the tech industry cannot self-regulate, and asks the committee to examine whether Navy Yard tenants align with its public mission.
Steve Kramer of Get Out the Vote says he sent an automated call mimicking the president's voice three years ago as a warning shot, argues the harms now seen were preventable, and urges the Council to think ahead because AI is already being governed by companies and bad actors may not be human.
Chair De La Rosa thanks the panel, excuses them, and moves to the Zoom panel with Sophie Cope, Ajit Singh, and Chaudhry Hassan Mahmood.
Sophie Cope says she spoke at the prior hearing about AI harms to women, calls SpaceX the poster child of such harms, warns that SpaceX can become a scapegoat for other AI labs, and urges the City Council to make New York City a beacon by restraining tech power.
Physician and public health practitioner Ajit Singh says AI enables corporations and agencies to exploit personal data, cites nonconsensual deepfakes, CSAM, sycophantic manipulation, and disinformation, calls SpaceX AI a bad actor, and says New Yorkers need enforceable safeguards and accountability.
Singh welcomes Introduction 1116's private right of action and Introduction 1101's reporting mechanism but says enforcement depends on definitions, evaluator independence, and court access; he notes the City Council tasked the administration to build the Office of Algorithmic Accountability by June and says New Yorkers should not endure preventable harms.
Department of Enterprise Technology employee Chaudhry Hassan Mahmood says AI tools are deployed by private contractors on FireCAD and EMSCAD emergency dispatch with PII and PHI, without AI governance policy, civil service review, or public disclosure, and asks for a public AI tool registry, FOIL-able records, controller audits, civil service sign-off, and whistleblower protections.
Chair De La Rosa acknowledges Council Member Wong has joined and asks anyone registered to testify who has not been called to use the Zoom hand raise function to be called in order.
Speaker Menin says SpaceX AI's answers were wholly unsatisfactory, citing Devin Kim's whistleblower testimony about his firing, the company's tasteful nudes standard, and the hyper-sexualized images of herself and thousands of women and children that Grok can generate within seconds; she thanks Kim for his bravery.
Chair De La Rosa thanks Speaker Menin and committee members, says the Council will continue moving AI legislation and discussing the Office of Algorithmic Accountability, thanks all who testified, and adjourns the hearing.