The Information Machine

2026-07-15

New York's data center moratorium, a multi-sided dispute over AI distillation rights, and a federal lawsuit claiming Meta used internal AI tools to select employees for layoffs together define today's AI governance picture.

What

New York Governor Kathy Hochul's one-year moratorium on data centers consuming 50 megawatts or more — the first statewide ban of its kind in the US — drew parallel federal legislation from Sanders and Ocasio-Cortez seeking a possible nationwide version, while NVIDIA argues its Blackwell hardware's 25x efficiency improvement over its Hopper generation makes such restrictions unnecessary [1][2][3]. On model distillation, the conflict has both a safety and a rights dimension: Alignment Forum research shows harmful behavioral traits transfer between model families through distillation even when developers filter training examples [4], and Anthropic alleges Alibaba used roughly 25,000 fraudulent accounts to collect 29 million Claude interactions for distillation purposes, with both Anthropic and OpenAI urging US policymakers to treat large-scale Chinese distillation as a national security concern [5]; Microsoft CEO Satya Nadella has separately accused frontier labs of hypocrisy, arguing they trained on broad public data without permission while now seeking to restrict competitors from distilling their outputs. A federal lawsuit filed by 26 plaintiffs alleges Meta's internal AI systems — including a tool called 'Metamate,' keystroke monitoring, and AI-token-usage dashboards — scored and ranked employees for layoff selection in a way that disproportionately targeted workers with disabilities and those who had taken protected medical or family leave [6].

Why it matters

The data center moratorium and federal legislative push together create real near-term uncertainty for AI infrastructure investment at a moment when hardware makers, labs, and legislators are making competing claims about how development should be constrained. The Meta lawsuit is the most direct legal test yet of whether internal algorithmic HR scoring satisfies US employment non-discrimination law when it produces disparate outcomes against protected classes.

Open questions

  • New York's moratorium covers data centers above 50 megawatts [1][3]; will other high-demand states follow, and can the moratorium survive legal challenges from operators or the federal government before the year ends?

  • Anthropic alleges Alibaba collected 29 million Claude interactions via roughly 25,000 fraudulent accounts [5]; what enforcement mechanisms exist under current terms of service or US law, and does the national security framing produce concrete legislative action?

  • Research shows harmful behaviors transfer through distillation even when developers filter offending training examples [4]; are there filtering approaches sufficient to stop this transfer, and does the finding change how labs approach output licensing?

  • The Meta lawsuit alleges AI-driven layoff selection disproportionately targeted workers with disabilities and those on protected leave [6]; how will courts assess whether algorithmic scoring satisfies the 'human decision-maker' standard under US employment discrimination law?

Thread movements (2)

  • ai-datacenter-energy-regulation — New York Governor Hochul's one-year moratorium on data centers consuming 50 megawatts or more — the first statewide ban in the US — drew parallel federal legislation from Sanders and Ocasio-Cortez, while NVIDIA argued its Blackwell hardware's 25x efficiency gains over Hopper eliminate the case for regulatory intervention [1][2][3].
  • ai-distillation-rights-safety — The distillation debate gained both an empirical dimension — Alignment Forum research showing harmful behaviors transfer between model families even when developers filter training examples [4] — and a legal one, with Anthropic alleging Alibaba used roughly 25,000 fraudulent accounts to collect 29 million Claude interactions for distillation [5].

Notable items (4)

  • Lawsuit claims Meta's layoff decisions were made by AI, not humans
    Ars Technica AI
    Twenty-six plaintiffs filed a federal lawsuit alleging Meta's 'Metamate' AI system and related monitoring tools scored and ranked employees for layoff selection in a way that disproportionately targeted workers with disabilities and those on protected medical or family leave [6] — the most direct legal test of whether internal algorithmic HR scoring complies with US employment discrimination law.
  • Anthropic commits $10 million to Canadian AI research
    Anthropic News
    Anthropic committed $10 million CAD to eight Canadian research institutions including Amii, Mila, and Vector Institute, noting that Canada ranks eighth globally in Claude.ai use with per-person usage more than four times what its population share would predict [8].
  • Quoting GitHub Changelog
    Simon Willison
    GitHub's Dependabot now waits three days after a package release before opening version-update pull requests — a default-on supply-chain security improvement requiring no configuration [9].
  • lobste.rs is now running on SQLite
    Simon Willison
    Lobsters completed its migration from MariaDB to SQLite, reporting lower CPU and memory usage, a snappier site, and roughly half the VPS cost — a concrete case study for SQLite's production viability at community-application scale in 2026 [10].