Home › Blog › The Maestro, Season 2 · Part 12
The Maestro, Season 2 · Part 12 · Phase 7 · 📜 CEO hat

Crossing the EU Line

Part 10 left Maestro-2 at 96% of the EU AI Act's compute line. A model is not trained in one stage, though, and by the time the full plan was added up the total was 3.7% over. This part is about a choice most start-ups would rather not face: shave the plan to stay under the line, or cross it on purpose and take on the obligations of a frontier provider.

FL
FrontierAGI Team
💼 CEO⚖️ Policy🛡️ Safety
⚠️ How to read this (as of 4 Nov 2026) Gopala's company is a teaching scenario, and nothing here is legal advice. eu_line.py and comply_plan.py were run (real output below). The 10^25 threshold and the obligations come from the AI Act's text (Articles 51–55); how each training stage is counted should be checked against the Commission's guidelines with counsel. Costs and timings are scenario estimates. Disclosure: Claude, made by Anthropic, helps write this site.

1. Month 36: three point seven percent

📖 The story

Hana Sato, the company's counsel, had a habit of reading spreadsheets from the bottom up. That was why she found it.

The pre-training line said 9.60e24. Everyone had seen that number; it was the one in the board deck, ninety-six percent of the line. Under it, in rows nobody had summed, were the long-context stage, the reinforcement-learning stage in the new environments, and a small allowance for customer skill fine-tunes. Hana added them up and wrote the total in red: 1.04e25. You are a systemic-risk model.

Gopala read it twice. Then he asked the only question that mattered: Is it cheaper to be under the line, or to be good?

🔬 In plain words: the EU compute line The EU AI Act sorts general-purpose AI models into two groups. All providers must document their models and respect copyright rules. Models trained with more than 10^25 operations in total are presumed to have "systemic risk" and their providers take on extra duties: evaluation, adversarial testing, risk mitigation, incident reporting and cybersecurity. The number is a presumption, not a verdict: the Commission can adjust it and can designate models on other grounds.

2. The map: stages add up, and the line triggers duties

🕸️ From a compute ledger to obligations

or click any node

Training stages in purple add up to the total (red). Crossing the line triggers four duties (green and blue).

3. Adding up the ledger

eu_line.py · Python 3, no dependencies; Article 51 threshold, scenario ledger
"""eu_line.py — will Maestro-2 cross the EU AI Act compute line?

The AI Act presumes a general-purpose model has systemic-risk capabilities when
its cumulative training compute exceeds 10^25 FLOPs (Article 51(2)). Which later
stages count, and how exactly, should be checked against the Commission's
guidelines with counsel. The ledger entries are SCENARIO estimates.
"""
LINE = 1e25
LEDGER = [("pre-training, option B2", 9.60e24), ("long-context mid-training", 0.30e24),
          ("RL post-training (environments)", 0.45e24), ("skill fine-tunes (per customer)", 0.02e24)]

def main():
    tot = 0
    for name, f in LEDGER:
        tot += f; print(f"{name:34} {f:9.2e}   running total {tot:9.2e}  ({tot / LINE:5.1%} of line)")
    print(f"\ntotal {tot:.2e} FLOPs: {'ABOVE' if tot > LINE else 'below'} the 1e25 presumption by {abs(tot - LINE) / LINE:.1%}")
    need = (tot - LINE) / (6 * 32e9)
    print(f"to stay under, cut pre-training by {need / 1e12:.1f}T tokens (from 50T) or drop the RL stage")

if __name__ == "__main__":
    main()
$ python3 eu_line.py # cumulative training compute pre-training, option B2 9.60e+24 running total 9.60e+24 (96.0% of line) long-context mid-training 3.00e+23 running total 9.90e+24 (99.0% of line) RL post-training (environments) 4.50e+23 running total 1.04e+25 (103.5% of line) skill fine-tunes (per customer) 2.00e+22 running total 1.04e+25 (103.7% of line) total 1.04e+25 FLOPs: ABOVE the 1e25 presumption by 3.7% to stay under, cut pre-training by 1.9T tokens (from 50T) or drop the RL stage

Cumulative training compute vs the 10^25 line

Pre-training (B2)96.0%
+ long-context mid-training99.0%
+ RL post-training103.5%
+ customer skill fine-tunes103.7%

Real output. Running totals, as a share of the line.

🔎 Go deeper: which stages count?
  • The Act speaks of cumulative compute used for training.
    • The Commission's guidelines for general-purpose AI providers (2025) explain how to estimate it; whether and how later fine-tuning and synthetic-data generation are included is exactly the detail to confirm against them with counsel ⚠️.
  • The team counted conservatively: every stage that changes the model's weights.
    • Counting conservatively costs little; being found under-counted later costs the company's credibility.

4. Shave or cross

📖 The story

Tomás wanted to shave. Two trillion fewer tokens and we are under. Nobody will notice the difference in quality.

Elena Brandt, who ran safety, wanted to cross. We already run most of this. Evaluations, red-teaming, incident response. The law is asking us to write down what we do.

Hana gave them both the same answer. If you stay under by 2%, every future stage you add, every retrain for a customer, puts you back over. You will be managing an accounting line instead of a model.

Gopala decided before the meeting ended. They would cross, on purpose, and say so publicly. A company whose product is evidence that changes are safe could not be the company that trimmed its run to avoid evaluation.

OptionComputeWhat it costsWhat it risks
Shave 1.9T tokens~1.00e25A slightly weaker modelAny later stage crosses anyway
Drop RL post-training9.92e24The stage that makes Maestro good at its jobProduct quality
Cross and comply (chosen)1.04e25$2.20M, 12 weeks in parallelA public commitment the team must keep

5. What crossing costs

comply_plan.py · Python 3, no dependencies; Articles 52–55, scenario costs
"""comply_plan.py — what crossing the line costs.

Obligations for providers of general-purpose models with systemic risk come
from AI Act Articles 52-55 (notify the Commission, documentation, evaluations
including adversarial testing, risk mitigation, serious-incident reporting,
cybersecurity). Owners, weeks and costs are SCENARIO estimates.
"""
OB = [("notify the Commission (Art. 52)", "CEO + counsel", 1, 0.02),
      ("update technical docs + downstream info (Art. 53)", "research + docs", 6, 0.15),
      ("update copyright policy + data summary (Art. 53)", "data lead + counsel", 4, 0.10),
      ("evaluations incl. adversarial testing (Art. 55)", "safety team + external red team", 10, 0.90),
      ("systemic-risk assessment + mitigation (Art. 55)", "safety lead", 8, 0.35),
      ("serious-incident reporting process (Art. 55)", "ops + safety", 3, 0.08),
      ("cybersecurity for weights and infra (Art. 55)", "security", 12, 0.60)]

def main():
    for name, who, wk, c in OB: print(f"{name:52} {who:32} {wk:3} wk  ${c:4.2f}M")
    tot = sum(o[3] for o in OB); crit = max(o[2] for o in OB)
    print(f"\ntotal ${tot:.2f}M; longest item {crit} weeks (runs in parallel with the ~12-week training run)")
    print(f"share of the Maestro-2 compute bill (~$19.5M on 4,096 GPUs): {tot / 19.5:.0%}")

if __name__ == "__main__":
    main()
$ python3 comply_plan.py # obligations, owners, weeks, cost notify the Commission (Art. 52) CEO + counsel 1 wk $0.02M update technical docs + downstream info (Art. 53) research + docs 6 wk $0.15M update copyright policy + data summary (Art. 53) data lead + counsel 4 wk $0.10M evaluations incl. adversarial testing (Art. 55) safety team + external red team 10 wk $0.90M systemic-risk assessment + mitigation (Art. 55) safety lead 8 wk $0.35M serious-incident reporting process (Art. 55) ops + safety 3 wk $0.08M cybersecurity for weights and infra (Art. 55) security 12 wk $0.60M total $2.20M; longest item 12 weeks (runs in parallel with the ~12-week training run) share of the Maestro-2 compute bill (~$19.5M on 4,096 GPUs): 11%

Compliance work, in weeks (parallel with training)

Cybersecurity for weights and infra12 wk · $0.60M
Evaluations incl. adversarial testing10 wk · $0.90M
Systemic-risk assessment8 wk · $0.35M
Update technical documentation6 wk · $0.15M
Copyright policy + data summary4 wk · $0.10M
Incident-reporting process3 wk · $0.08M
Notify the Commission1 wk · $0.02M

Real output. Total $2.20M, about 11% of the compute bill.

6. Sandbox: the compute ledger

🎮 Sandbox: the compute ledger

Same ledger as eu_line.py (32B active parameters, 6ND, plus 0.02e24 for customer skill fine-tunes). Not legal advice: how each stage counts should be checked against the Commission's guidelines.

7. Five mistakes

8. Delegation box

🙋 Gopala himself
  • The decision to cross, and saying so publicly
🤖 AI agents
  • Compute ledger tracking
  • Drafting documentation from the run records
🤝 Bought, rented or borrowed
  • Counsel for the notification
  • An external red team for adversarial testing

9. Budget and risk tracker

ItemStatus after Part 12Note
Compute1.04e25 planned3.7% over the line
StatusSystemic-risk GPAI providerCommission notified (story)
Compliance$2.20M · 12 weeksin parallel with training
CredibilityStrengthenedcrossed on purpose

10. The series

The Maestro, Season 2: 23 parts in five phases (Season 1: 18 parts)

VerdictMaestro-2 crosses the EU compute line by 3.7% once every training stage is counted, and the company chooses to cross deliberately rather than shave the model to fit a number. The duties cost about 11% of the compute bill, mostly in evaluations and securing the weights. Next: the new ingredients that go into the run.

11. Reading the numbers

Primary / standard
Regulation (EU) 2024/1689 (the AI Act), Articles 51–55; the General-Purpose AI Code of Practice (July 2025).
Secondary ⚠️
Article summaries from artificialintelligenceact.eu ⚠️ (secondary).
Our assumptions
The ledger, costs, timings and all story events are the scenario. Not legal advice.