// regulation/policy

All signals tagged with this topic

Why Flock's Surveillance Model Represents a Dangerous Dead End

Rick Wilson argues that Flock's automated license plate reader network—marketed as crime-fighting infrastructure but operating as mass surveillance—has become indefensible as a business model and civic tool. The company's documented inability to prevent abuses (false arrests, stalking, harassment) while resisting meaningful oversight suggests some technologies cannot be reformed through better policies; they require shutdown. The case challenges the Silicon Valley playbook of "move fast, apologize later" by showing that some infrastructure projects create irreversible social damage before regulation catches up, making preventive opposition more rational than remedial governance.

AI-drafted bills create more work for House Legislative Counsel

Congressional staffers are discovering that AI-generated legislation requires so much editing it defeats the efficiency premise entirely. The Legislative Counsel's office now spends more labor hours correcting machine-drafted bills than writing them from scratch. This reflects a common pattern in institutional AI adoption: tools get deployed before anyone measures whether they actually save time, creating a hidden cost that undermines the case for deployment. Legislative drafting, despite seeming like an ideal AI task—formulaic, precedent-heavy, rule-bound—depends on tacit knowledge about political feasibility, unintended consequences, and constitutional edge cases that current language models don't capture.

EU mandates five years of phone updates; automakers face no equivalent rule

The EU's Digital Products Act requires smartphone makers to supply security and functionality updates for five years minimum, creating a legal floor for device longevity that the auto industry has successfully avoided despite cars lasting far longer and containing increasingly networked systems. This regulatory gap reflects how different sectors have negotiated with policymakers—tech companies absorbed the update burden as a compliance cost, while automakers used fragmented supply chains and safety liability arguments to sidestep similar obligations, leaving owners of decade-old vehicles running on obsolete software. As cars become computers, manufacturers can now engineer planned obsolescence through software while claiming zero legal responsibility for security vulnerabilities in systems that control braking and steering.

OpenAI's Crawler Ignores Robots.txt Rules for Training Data

OpenAI is allowing its GPTBot to bypass robots.txt directives that publishers use to prevent automated access, treating the standard as advisory rather than binding. This escalates friction between AI labs and content creators. Publishers lack technical recourse to stop training scrapes and must rely on legal action, shifting power to AI infrastructure companies that can unilaterally decide which rules apply to them.

Man Hides AI Prompts in Court Filings to Trick Judge

A litigant in Kansas attempted to embed hidden instructions in court documents—a technique called prompt injection—betting that a judge using AI tools to review filings would execute those commands instead of reading the actual legal argument. The gambit failed immediately because the judge was reading manually, but it exposes a real vulnerability: as courts adopt AI to manage document review and legal research, adversaries will exploit that automation layer with techniques designed to hijack the AI's outputs rather than persuade the human decision-maker. Courts will need to establish explicit protocols around AI use in adjudication before this becomes a common tactic.

Self-Represented Litigant Embeds AI Manipulation Code in Court Filing

A Connecticut pro se defendant inserted prompt injection attacks into their legal filing, attempting to hijack any AI system a judge might use to summarize or analyze the case. This is the first documented instance of adversarial prompt engineering deployed in actual litigation. The attack exposes a gap: courts are adopting AI to manage document volume faster than they are securing it.

Open-Source Models Already Won the AI Governance Debate

The compliance frameworks from OpenAI, Anthropic, and their Western peers are functionally irrelevant for the vast majority of deployments. Cheaper open-weight models from China have ensured that regulatory efforts focused on controlling leading labs address yesterday's chokepoint. Governance designed around monopoly protection doesn't constrain a fragmented market.