The frontier labs are asking city hall to regulate them before they build something they can't unbuild.
The Summary
- Executives from Anthropic, OpenAI, Meta, and Microsoft will appear before NYC lawmakers next week for a hearing on AI oversight
- The companies are specifically urging scrutiny of self-improving AI systems, systems that modify their own code and capabilities
- This marks a rare moment of industry consensus: the builders want guardrails before the next capability jump
The Signal
The frontier AI labs are doing something unusual: asking for regulation before they're forced to. Anthropic, OpenAI, Meta, and Microsoft executives will testify before New York City Council next week, specifically calling for oversight of self-improving AI systems. Not general "AI ethics." Not vague "responsibility frameworks." Self-improving systems. The ones that rewrite their own code.
This is not altruism. This is strategic positioning. When labs that compete for talent, capital, and model benchmarks suddenly agree on policy language, they've seen something in their evals that spooked them. Self-improving systems represent a capability threshold where the gap between "we shipped this" and "we understand this" gets uncomfortably wide.
"The builders want guardrails before the next capability jump."
NYC lawmakers are responding to growing concern about lack of AI guardrails, but city-level AI regulation is a strange venue for this fight. New York can't set technical standards for recursive self-improvement. It can't audit model weights or eval protocols. What it can do is create liability frameworks, procurement requirements, and disclosure mandates that make it expensive to deploy systems the city deems unsafe.
The timing matters. We're in the window between GPT-4 class models (impressive, occasionally useful, frequently wrong) and whatever comes next. If the labs are pre-negotiating the rules for self-improving systems now, they expect to ship them soon. Not in five years. Soon enough that city council hearings are worth their executives' time.
The Implication
Watch what regulations the labs don't ask for. They're calling for oversight of self-improvement, but notice what's missing: compute thresholds, open-weight restrictions, or mandatory capability disclosures. This is a negotiation. The labs are offering the city a framework that looks like oversight but leaves their core development pipeline untouched. If you're building on these platforms, the question isn't whether regulation is coming. It's whether the regulation will actually constrain the capability race or just formalize it.