Dieter Van Esch

The through-line

What recurs across Dieter Van Esch's public thinking is a preference for testing systems from the inside rather than arguing about them from the outside. He built and ran the fake celebrity investment ads himself to prove how easily they pass, spending a deliberately trivial sum to make the point: "Ik heb het met €150 budget gedaan om deze case te kunnen aantonen... Maar wat had ik met €150.000 kunnen aanrichten?" [2] The same instinct shows in how he describes sitting in on a two-day referees' training session as VAR operator, watching an institution wrestle with the tension between consistency and individual judgement [1].

The second constant is his distinction between inability and unwillingness. Whether the subject is a platform that lets fraud through or a group of officials trying to decide alike, his conclusion is that the outcome is chosen, and that the mechanism for changing it is uncomfortable, repeated, explicit conversation about expectations [1][2].

On Meta as an accomplice rather than a bystander

Van Esch does not treat scam advertising as a leak in an otherwise sound system. He calls it a business with a sponsor: "Dus dit is gewoon een industrie en die industrie wordt door, ik ga het luid op zeggen, meta ondersteund." [2] His sharpest evidence is what happened when he uploaded fake celebrity imagery for the ads: Meta's own AI came back with 20 improved variations, including turning Wim De Vilder's face into women of different ages so the campaign could reach new audiences [2]. His summary of that moment is blunt: "Een fake beeld gaan we nog faker maken en Facebook helpt hierbij." [2] The platform is not merely failing to intervene, it is optimising the fraud, and the felt experience of running these campaigns is one of encouragement: "Aan de andere kant heb je echt wel Meta die je aanmoedigt van blijf maar fake advertenties doen, vriend." [2]

On moderation as a governance choice

The claim that this cannot be stopped does not survive his own control experiment. Of roughly 50 fake celebrity scam campaigns, 20 went live without a hitch, some were briefly online before being rejected, and rejected ones sailed through after small tweaks, which leaves no discernible enforcement logic at all [2]. The decisive comparison is with the competition: LinkedIn and Google blocked the same test ads that Meta waved through [2]. Effective moderation is therefore demonstrably possible, and the difference is a matter of not wanting to rather than not being able to, on a platform earning in the region of $16B from fraudulent advertising, with 1,100 Bart De Wever scam ads and more than 4M views since January 2025 [2].

On how the scam funnel actually works

His account of the mechanics is what makes the scale intelligible. The entry point is often something harmless-looking, a fake prize contest that captures your data. From that moment you are flagged as susceptible and fed back through Meta's custom audiences, retargeted again and again with progressively more advanced scams [2]. The victim is not caught once, they are enrolled. And he is pessimistic about the arms race that follows: "De vraag is altijd wie is het snelst, de stroper of de boswachter? En helaas zal de stroper zich altijd blijven opnieuw uitvinden." [2]

On the ethical compass being the only thing in the way

The most uncomfortable part of his testimony is what the exercise revealed about himself. Having proven how frictionless the process is, he draws the conclusion without softening it: "Ik ben echt zeker dat ik heel snel mensen zou kunnen oplichten. Moest ik daar mijn job van maken, ik zou heel snel veel centen kunnen gaan verdienen. Mocht ik geen etisch kompas hebben." [2] The barrier protecting the public is not technical, contractual or algorithmic. It is the personal restraint of people who happen to have marketing skills and choose not to use them this way.

On uniformity that leaves room for personality

On consistency in decision-making he refuses the false choice between everyone deciding identically and everyone deciding as they please. Uniformity is what everyone says they want, and at the same time people want officials with room for their own personality: "Geen robots." [1] His answer to how both can hold at once is borrowed from a 2008 KBC Bank & Verzekering slogan: praten werkt [1]. In practice that means hammering on expectations and holding proper discussion about them, working through examples of both good and bad decisions, and accepting that the destination is shared while the route is not. As he puts it, allen naar Rome, maar iedereen via eigen weg, met eigen voertuig [1].

Takeaways

• Treat scam advertising on Meta as an industry with a backer, not an oversight: "die industrie wordt door, ik ga het luid op zeggen, meta ondersteund" [2].
• Meta's AI actively improved uploaded fake celebrity images, offering 20 variations and swapping Wim De Vilder's face for women of different ages to open up new target audiences [2].
• Ad approval showed no enforcement logic: of ~50 fake campaigns, 20 ran flawlessly, some were pulled after going live, and rejected ones passed after minor tweaks [2].
• LinkedIn and Google blocked the identical test ads Meta approved, which makes Meta's failure a governance decision rather than a technical limit [2].
• Once your data is captured through something like a fake prize contest, custom audiences mark you as susceptible and retarget you with escalating scams [2].
• A €150 budget was enough to demonstrate the whole case, which raises the obvious question of what €150,000 would do [2].
• Consistency in decision-making comes from repeated conversation about expectations and joint review of good and bad examples, not from turning decision-makers into robots [1].