DraftKings uses AI to target problem gamblers; 160,000 Germans petition against AI energy waste. Compliance theatre is everywhere this week.
elcome to issue twenty-two. This week we have a gambling company using AI to find the people most likely to lose money, and 160,000 Germans telling their parliament that AI data centres are eating the climate alive. Both stories are about who pays the hidden cost of AI optimisation, and neither cost appears on a vendor's compliance checklist. We dig into DraftKings' targeting engine and the German petition, plus what an LLM-free compliance tool on Hacker News tells you about where the smart builders are heading.
Let’s go.
yours, Flux
Flux Weekly is a 6-minute briefing for people who have to actually make AI work in Europe. Sole traders to enterprise, one issue every Friday morning.
- New We added a vendor-targeting checklist to the compliance scan this week, prompted by the DraftKings story, so you can flag whether a recommendation system's optimisation objective is compatible with your AI Act obligations.
- Updated The Flux resource library now includes the Medium practical guide on making AI agents EU AI Act compliant, which surfaced on Hacker News this week with useful agent-specific framing.
- ICYMI If you missed it: the OpenComplai team published a blog post explaining why their compliance engine uses zero LLM calls and why their CI pipeline blocks anyone from adding one, worth ten minutes of your time if you are building compliance tooling.
DraftKings Is Using AI to Find the Gamblers Most Likely to Lose, and That Should Alarm Every EU Operator Procuring Recommendation Systems
The algorithm knows you better than you know yourself. DraftKings has built an AI system that identifies which customers are most likely to place losing bets and respond to gambling promotions, then serves those people targeted advertising. According to EFF, this is a textbook case of behavioural advertising turbo-charged by data scale. The more data DraftKings holds on a user, the more precisely it can identify vulnerability and act on it.
This is not a fringe use case. Any EU operator running a recommendation or personalisation system, whether for e-commerce, media, or financial products, is operating machinery that functions on the same logic. The EU AI Act's provisions on prohibited practices and high-risk classification exist precisely because this kind of targeting, when applied to vulnerable groups, crosses a line regulators have already drawn. DraftKings just made that line impossible to ignore.
Does your AI inform a decision that affects a person's job, credit, education, or essential service?
- ✓AlgorithmWatch delivered 160,000+ signatures to the Bundestag demanding the government stop weakening energy-efficiency rules to favour AI data centre expansion.
- ✓AlgorithmWatch testified before the Bundestag's Interior Committee, calling on parliament and the Bundesrat to reject a proposed surveillance package backed by the interior and justice ministries.
- ✓The EU Commission published a draft Kids Act that EFF warns will put online services behind blanket age gates, expand intrusive age verification, and undermine privacy for all users, not just minors.
- ~DraftKings was exposed by EFF for deploying AI to identify and target the users most likely to place losing bets, a practice with direct parallels to prohibited AI uses under EU law.
- 1OpenComplai: No LLM in Compliance Decisionsengineering
A blog post by the OpenComplai team explaining why they built their EU AI Act compliance engine without a single LLM call and locked that decision into their CI pipeline.
Why we like it. It is a rare honest account of the engineering tradeoffs in compliance tooling, and it asks a question every vendor should answer: why would you let an unpredictable model decide whether you are compliant?
- 2Practical Guide: AI Agent EU AI Act Complianceagents
A Medium post by Mir Arshad Talpur walking through how to make an AI agent compliant with the EU AI Act, covering transparency, human oversight, and risk classification.
Why we like it. Agent-specific guidance is still rare; most compliance writing assumes a static model rather than an autonomous system taking actions on your behalf.
- 3
The hidden cost is always the objective function
By John Ferguson
Every AI system has an objective function. It is the thing the model is trying to maximise. Most vendor contracts do not mention it. Most procurement checklists do not ask for it. And yet it is probably the single most important piece of information you need to decide whether a system is safe to deploy.
DraftKings did not set out to harm gamblers. They set out to optimise revenue. The harm is a downstream consequence of the objective, not a bug in the code. That distinction matters legally, but it does not make the outcome better for the person on the receiving end.
The EU AI Act is, at its core, a law about who bears the cost of optimisation. When an algorithm maximises profit by finding vulnerable people, the cost lands on those people. The regulation asks operators to take that cost seriously before they deploy, not after a regulator comes knocking.
My practical advice this week: pull up the documentation for every AI system you are running or procuring and find the sentence that describes what it optimises for. If that sentence does not exist, that is your first gap. The other 425 days before high-risk rules apply will go faster than you think.
John Ferguson · Founder, Agentic Fluxus
Short answer.Start by asking your vendor one question: what is the objective function this model optimises for? If the answer is 'conversion' or 'revenue per user' without any constraints on user vulnerability, you have homework to do. Under the EU AI Act, systems that exploit behavioural weaknesses sit very close to prohibited territory. Get the answer in writing, document it, and review it against Article 5's prohibited practices list.
When you last reviewed a vendor's AI system, did you ask what the model is actually optimised to maximise?
The Trump administration has been routing official government announcements through Truth Social's paid early-access tier, which charges organisations up to $100,000 a month to see posts first. EFF filed a federal court brief this week arguing that blocking equal access to official government statements violates the First Amendment.
DraftKings' AI system does not just serve ads. According to EFF's analysis, it specifically identifies customers most likely to place losing bets and most likely to respond to gambling promotions, then targets them. The product optimises for the customer's failure.