AI Policy, Follow the Money: A Timeline of Fines, Funds, and Compliance Costs
Archi4 min read
Contents
- The money arrived before the rules did
- April 2021: Brussels picks a number
- September 2022: China's algorithm registry gets a price tag
- October 2023: The G7 code of conduct, unpaid
- November 2023: The UK Safety Institute opens with 100 million pounds
- March 2024: The EU AI Act passes, and compliance firms smell blood
- May 2024: Colorado passes a law with a litigation clause
- August 2024: California's SB 1047 becomes a funding fight
- February 2025: The Paris summit and the 400 million euro question
- Mid-2025: The first AI Act fines are still hypothetical
- The quiet costs
- What to watch next
The money arrived before the rules did#
When the European Commission published its first draft of the AI Act in April 2021, the document ran to 108 pages. The fine print drew less attention than the headline number: penalties of up to 6 percent of global annual turnover for the most serious violations. That figure set a benchmark. Every subsequent AI policy debate has been, in part, an argument about who pays and how much.
This timeline tracks the money: fines levied, funds announced, compliance budgets inflated, and the quieter costs that never make a press release.
April 2021: Brussels picks a number#
The draft AI Act proposes a tiered penalty structure. Prohibited practices draw up to 30 million euros or 6 percent of worldwide turnover, whichever is higher. Other breaches top out at 3 percent or 15 million euros. For a company like Google, with revenue north of 280 billion dollars, that 6 percent figure is theoretical but instructive: no regulator has ever collected it, and no compliance officer wants to test it.
September 2022: China's algorithm registry gets a price tag#
China's rules on recommendation algorithms require companies to register their systems with the Cyberspace Administration. By late 2022, the registry lists hundreds of entries. The compliance cost is mostly unglamorous: documentation, audits, and the engineering hours to modify ranking systems. Industry surveys put the internal cost for a mid-sized platform in the low millions of yuan, a rounding error for ByteDance but a real line item for smaller firms.
October 2023: The G7 code of conduct, unpaid#
G7 leaders announce a voluntary code for advanced AI developers. It has no fines, no enforcement body, and no budget. The money angle is the opposite of Brussels: companies pledge to spend on red-teaming and evaluation, but the commitments are self-reported. Analysts note that voluntary frameworks tend to cost less upfront and more later, when mandatory rules arrive and firms must retrofit documentation they never kept.
November 2023: The UK Safety Institute opens with 100 million pounds#
The UK government funds its AI Safety Institute with roughly 100 million pounds over two years. It is one of the first public budgets dedicated to testing frontier models. The institute's early work focuses on pre-deployment evaluations. The spending is small compared with the models themselves, where training runs can cost tens of millions of dollars, but it establishes a public buyer for safety research.
March 2024: The EU AI Act passes, and compliance firms smell blood#
The European Parliament approves the AI Act. Fines remain as drafted. Within weeks, consultancies publish rate cards for AI Act readiness assessments. Law firms add practice groups. A compliance industry forms before the first enforcement action, which is itself a kind of cost: companies pay now for rules that phase in over two to three years.
May 2024: Colorado passes a law with a litigation clause#
Colorado enacts a broad AI consumer protection law, later amended. The original text includes a right of action for violations, which turns compliance into an insurance question. Insurers begin asking developers whether they have documentation for training data and bias testing. Premiums for AI liability coverage remain hard to price, and several underwriters decline to offer policies for generative systems altogether.
August 2024: California's SB 1047 becomes a funding fight#
California's proposed frontier model law draws more than 100 million dollars in lobbying and campaign spending, according to state filings, with both industry groups and safety advocates pouring money into the debate. Governor Newsom vetoes the bill in September. The episode shows that the cost of AI policy includes the cost of shaping it, and that number is measured in the tens of millions before any rule takes effect.
February 2025: The Paris summit and the 400 million euro question#
The AI Action Summit in Paris produces pledges of public investment in AI infrastructure, including a French commitment of roughly 400 million euros for AI research and compute. The private sector announcements run higher: several hundred billion dollars in data center pledges globally, much of it already planned. The gap between announced investment and money actually deployed is the recurring story of AI economics.
Mid-2025: The first AI Act fines are still hypothetical#
By mid-2025, the EU AI Act's prohibited practices provisions have applied for months, but no public fine has been issued. Enforcement bodies are still staffing up. Experts estimate that national market surveillance authorities in several member states have fewer than ten dedicated AI staff each. The fine schedule exists on paper; the collection mechanism is a work in progress.
The quiet costs#
The visible numbers, fines and funds, dominate headlines. The larger sums sit in compliance departments, legal reviews, and engineering time. A 2024 survey of enterprise AI teams found that companies spend an average of 15 to 20 percent of AI project budgets on governance and documentation. For a startup, that can mean a full-time hire before the first customer.
What to watch next#
The EU's general-purpose AI rules take full effect in August 2025, with obligations for model documentation and copyright policy. The first enforcement actions, whenever they come, will set the real price of noncompliance. Until then, the money in AI policy is mostly being spent on preparation, lobbying, and the insurance industry's attempt to price a risk it does not yet understand.
More articles