Author: Luiza Jarovsky Source: https://www.luizasnewsletter.com/p/the-rise-of-technical-ai-policies Published: July 7, 2025 (edition #216)
"The Rise of Technical AI Policies" is a July 7, 2025 newsletter essay (edition #216) by Luiza Jarovsky. It argues that the AI governance community overestimates the role of law in establishing governance frameworks, and that the field also needs technical AI policies — direct technical interventions, certifications, and infrastructure tools that operate on shorter time horizons than law. Full framework coverage is at Technical AI Policies.
Summary of argument
Jarovsky argues that law is essential but limited as a governance tool, citing three structural constraints: formal-procedural slowness; its interpretive nature, in that humans interpret, operate, and apply it; and its constant exploitation by people, companies, and governments. On that basis she contends that, beyond regulation, the field needs technical AI policies that operate on shorter time horizons than legal processes.
The essay characterizes technical AI policies as a distinct category of governance intervention that complements law rather than replacing it. Jarovsky also names what she calls a democratizing factor: interdisciplinary technical-policy work empowers more people and social groups beyond those with specialized legal training.
AI copyright worked example (Summer 2025)
Jarovsky uses AI copyright as a worked example of technical interventions moving faster than court decisions. In summer 2025, two US courts sided with AI companies in copyright training-data cases: Anthropic received a partial fair-use win, and Meta received a fair-use win in which the judge explicitly stated that most AI training infringes copyright. She describes this as a major blow for content creators who had expected legal acknowledgment.
Against that, she notes that six days before those rulings, Cloudflare announced new tools letting content creators control whether AI bots can access their websites. Regardless of what the law says or how it is enforced, that globally available technical mechanism lets publishers opt out of AI training. She gives Fairly Trained as a second example: a voluntary certification that certifies AI companies that license their training data, signaling respect for creator wishes regardless of legal mandates.
Hidden-prompts example
The essay's second example draws on a July 2025 Nikkei Asia report that scientific papers from 14 academic institutions in 8 countries contained hidden AI prompts instructing AI systems to give positive reviews. Jarovsky notes that some researchers defended the practice as a counter-measure against AI-using reviewers. The technical AI policy she identifies in this context is technical mechanisms throughout the peer-review process to detect and block hidden AI prompts, observing that a legal reaction would take years.
Forms technical AI policies take
Jarovsky identifies several forms technical AI policies can take, treating detection-and-blocking as a category separate from infrastructure tools and certifications:
- Infrastructure tools (Cloudflare-style content controls, watermarking, content provenance)
- Certifications (Fairly Trained-style training-data licensing certification)
- Detection and blocking mechanisms (hidden-prompt detectors, jailbreak filters, agent identifiers)
- Voluntary best-practices (lab RSPs, model cards, system cards)
- Interdisciplinary process interventions (publication review, conference vetting, third-party evaluation)
Relationships
- authored-by: Luiza Jarovsky
- introduces: Technical AI Policies
- related: Anthropic's Responsible Scaling Policy (Version 3.1), System Card Due Diligence, AI Content Provenance, AI Content Licensing
- contradicts: law-only framing of AI governance critique