Seattle makes algorithmic grocery pricing illegal
Seattle just became the first US city to ban the practice, and Google quietly admitted its new private AI memory gives up a privacy guarantee it used to make.
The Seattle City Council voted 7-2 on Tuesday to ban personalized grocery pricing — the first US city to do it. The Fair Pricing and Transparency Act (CB 121267) forbids using a shopper's personal data to set the price they see for groceries and other essentials: browsing history, real-time location, inferences about income, family size or health conditions. Prices must be posted and available to everyone equally. The ordinance names specific retailers suspected of the practice, and applies to large grocery operations with 20 or more locations plus delivery platforms. Councilmembers Kettle and Rivera voted no; Mayor Katie Wilson, who championed the bill, called it a matter of equal treatment — "People have been clear: they don't want their data fed into algorithms that decide how much they pay." The bill now goes to her desk for signature.
The evidence behind it is the part worth keeping. Consumer Reports ran nearly 400 shoppers buying the same basket at the same time and found Instacart quoting different prices for the same product from the same store — differences reaching 23% on some items, worth more than $1,200 a year to a family. Instacart ended that test program after the investigation, but told CR it would still let retail partners test promotions through the platform. Maryland, Connecticut and New Jersey already passed state-level bans. Seattle is the first city to try it, which matters more than the ordinance's narrow scope: cities can move faster than Congress, and grocery pricing is the most legible version of a problem that applies to every algorithmically-priced thing you buy. Our take: the target is right and the instrument is crude. A ban on personalized pricing still permits the practice that actually costs shoppers money — discounting off an inflated list price — which is exactly what we argued in The Take — Give shopping agents a name, not a permission slip: transparency about the mechanism, not just the price, is the missing piece.
Google says it can now give Gemini persistent memory across your devices while keeping the keys on your phone — and in the same breath it drops a privacy promise it used to make. In a technical update, the Google Private AI Compute team described a new server-side memory layer: your context lives encrypted in per-user databases inside hardware-isolated enclaves on AMD SEV-SNP, the decryption keys are bound to your device and never visible to Google, and records are decrypted only in volatile enclave RAM for the length of a request. It's the fix for a real problem — the platform was strictly stateless until now, wiping everything when a task ended, which is why assistants couldn't remember anything worth remembering.
The honest part is a line in the updated technical brief: because a request has to reach your store, the system must resolve a stable per-user identifier, so "the stateful system does not claim network-level non-targetability." The stateless version did claim that, using third-party IP-blinding relays to make targeting one user's traffic infeasible. Persistent memory is inherently targetable — that is a genuine trade, stated plainly rather than buried. Google also published an independent audit by Trail of Bits, whose 54-page assessment found ten issues: two High, one Medium, two Low, five Informational, with two left unresolved. The open High-severity one concerns deterministic hashes of model response sub-strings being sent outside the trusted boundary for recitation checks — Google argues no raw prompts or outputs leave, that exploitation needs privileged internal access, and that stronger protections are being implemented. Memory does not ship yet, and the brief carries no date. Worth noting what the stateless platform promised before: the memory-enabled version is where the guarantee gets thinner, and Google is the one saying so.
What to watch: whether Mayor Wilson signs without amendment, and whether Google ships the memory layer with the two unresolved audit findings closed.
Would you trade a targeting guarantee for an assistant that actually remembers you — or is persistent memory the line you don't cross? Tell us in the comments.
Sources: Consumer Reports · FOX 13 Seattle · Seattle City Council · Google DeepMind · Google Private AI Compute technical brief (PDF) · Trail of Bits security assessment (PDF) · Unite.AI
