The Brief · Issue 8

Week of 17 August 2026 — the cost of letting the algorithm decide

Uber was fined €825M for automated driver deactivation, the FTC proposed personalized-pricing rules, and ANA moved on retail media measurement.

September 22, 2026

Compiled on 22 September 2026 for The Brief's launch archive, from sources published between 17 and 23 August 2026.

Four stories that mattered

Dutch regulator fines Uber roughly €825 million over automated driver deactivation

The Dutch data protection authority fined Uber approximately €825 million (about $966 million), reported 21–23 August, for automatically deactivating driver accounts — in some cases permanently — based on algorithmic fraud or low-rating signals, without adequate human review. Deputy chair Monique Verdier said "a computer should not make decisions on its own that have [such] major consequences." It is the second-largest GDPR fine on record, after Meta Ireland's €1.2 billion. Uber said it "strongly disagree[s]" and plans to appeal.

This is the clearest recent example of GDPR Article 22's limits on fully automated decisions being enforced at scale against a live production system, not a hypothetical — relevant to any team running an unreviewed scoring or eligibility model that acts on EU residents.

Source: TechCrunch, "Uber faces fine of nearly $1B over automated driver suspensions"

FTC proposes enforcement policy on personalized pricing

The Federal Trade Commission published a proposed enforcement policy statement on personalized pricing on 19 August, saying retailers that use personal data to set individualized prices while implying a static price may violate the FTC Act's ban on unfair or deceptive practices. Chairman Andrew Ferguson said "businesses that fail to tell consumers how their personal data is being used to set a price may be in violation of the FTC Act." The commission voted 2–0 to publish, with public comments open through 18 September.

Combined with this month's Senate hearing on the same subject, personalized pricing has moved from an academic complaint to an active target of both a Senate committee and the FTC in the space of three weeks — a trend worth flagging to anyone building dynamic-pricing or personalization models.

Source: Federal Trade Commission, "FTC Seeks Comment on Enforcement Policy Statement Regarding Personalized Pricing"

ANA pushes for a shared standard in retail media measurement

The Association of National Advertisers published its first framework for standardizing retail media measurement on 18 August, urging marketers to rely more on independent third-party validation — recommending Media Rating Council baseline standards for impressions, viewability, clicks and invalid traffic — rather than each retail media network's self-reported numbers. The framework's longer-term goals include incrementality measurement and cross-network data sharing.

Retail media is one of the largest and least standardized parts of the ad-measurement stack; a shared vocabulary from the industry's own advertiser trade body is a concrete step toward analytics teams being able to compare retail media networks the way they already compare search or social channels.

Source: Marketing Dive, "ANA cautions marketers against overreliance on retail media data"

Snowflake adds an AI mode to sensitive-data classification

Snowflake released, in public preview on 17 August, an AI mode for its sensitive-data classification feature, which uses a language model to identify additional semantic categories of sensitive data beyond what pattern-based classification catches on its own. It can be turned on through the Trust Center's classification profile settings or via SQL.

Using an LLM to help find sensitive data is a useful capability, but it changes the error profile of classification from missed pattern matches to model misjudgment — worth validating carefully before it drives access-control or masking decisions automatically, in the same week a regulator fined a company for exactly that kind of unreviewed automated decision.

Source: Snowflake, "AI mode for sensitive data classification (Public Preview)"

Tool moves

One how-to

Run the Article 22 test before a model gets to decide anything on its own. The Uber fine turned on a specific, narrow question: was a decision "based solely on automated processing" with a "legal or similarly significant effect," and if so, did the person get a meaningful way to contest it? A short checklist for any scoring, deactivation, eligibility or pricing model that touches EU residents:

  1. Name the decision, not the score. A risk score alone isn't the issue; what happens automatically because of that score is. List every action your system can take without a person signing off.
  2. Check for "solely automated." If a human reviews and can overrule the output before it takes effect, the strictest Article 22 rules don't apply — but that review has to be real, not a rubber stamp on an interface that makes overriding difficult.
  3. Weigh the effect. A "significant effect" covers things like account suspension, credit denial or pricing — not every automated action. Rank your model's possible actions by how hard they'd be to reverse.
  4. Build in the right to contest. Anyone the system acts on needs a real path to ask for a human review and get an explanation — which means keeping the inputs and logic auditable, not just the output. An explainable AI approach that can produce a specific reason, not just a confidence score, makes this much easier to satisfy.
  5. Document it as you would any other control. Treat this as part of ordinary data governance, with the same audit trail expectations as role-based access control, so the answer to "why was this account deactivated" exists before someone asks.

Getting this wrong doesn't just risk algorithmic bias — as Uber shows, it can be a fineable offense on its own, regardless of whether the underlying decision was accurate.

One number

$72 billion — Emarketer's forecast for U.S. retail media ad spending in 2026, up 19% from 2025, cited in the Association of National Advertisers' 18 August push for standardized retail media measurement.

The figure is a forecast, not a measured total, and it aggregates spend across retail media networks that the ANA's own announcement says use inconsistent metrics and methodologies — so it says how much money is flowing into the channel, but not how comparably that spend is being measured or how much of it retailers can attribute to incremental sales.

Get the next one