Recall Forensics

Recall Forensics: The Darwin's Cat Food Salmonella Advisory — the validation failure behind it

Team ValidLeaf · November 2026 · 6 min read

Recall Forensics is our series tracing real product recalls and enforcement actions to the specific validation gap behind them. Every fact comes from public regulator records or reporting of those records — no speculation, no invented causes. This case starts somewhere no quality system should ever start: with two sick cats, one of them hospitalized. Only after a consumer complained did testing find Salmonella in two lots of a raw cat food — a product with no cooking step, no expiration date on its label, and a lot code printed on just one of the four units in each package.

What happened

On October 26, 2026, FDA cautioned pet owners not to feed two lots of Darwin's Natural Pet Products Natural Selections Chicken Recipe for Cats, made by Arrow Reliance, Inc. The agency said two samples of the food had tested positive for Salmonella. FDA collected and tested those samples after receiving a consumer complaint about two cats that developed gastrointestinal illness after eating the food. One cat was hospitalized after eating food from one of the two affected lots, and both cats recovered after they stopped eating the products (FDA advisory, October 26, 2026).

The affected food is Lot 12731 (MFG Date: Jul 08, 2026) and Lot 12493 (MFG Date: Apr 02, 2026) — two lots manufactured more than three months apart. It was sold online through a subscription service, frozen, in 2-pound white and clear plastic packages with green and black labeling, each package split into four separate units. The lot code is printed on the front of the lower left unit only. The labels carry no expiration date, so FDA warned the food could still be in consumers' freezers and told owners to check before feeding. Owners with an affected lot were told to throw it away in a secure container, not donate it, and to clean and disinfect supplies and surfaces it had touched. If the part of the package carrying the lot code is gone, or the code cannot be read, FDA's instruction is to throw the food away anyway. The agency also notes a fact that matters for everything below: neither freezing nor refrigeration kills Salmonella (FDA).

FDA said it recommended that Arrow Reliance recall the two lots, and that, as of the advisory, the firm had not initiated an adequate recall of the affected products. Food Safety News, reporting on October 27, described the firm as having refused to initiate a recall (Food Safety News). A later listing changed the form, if not the visibility, of that picture: eFoodAlert's November 3–4 roundup of recall notices reported an Arrow Reliance recall entry for the same two lots, noting that no public recall notice was issued, consignees were notified by email, and the total quantity of recalled product was unknown at that time (eFoodAlert, November 4, 2026).

For balance, the company's position is also on the public record. Reporting by dvm360 notes that Darwin's disputes the allegations in the separate federal case against it, that its owner has called FDA's zero-tolerance policy for Salmonella in raw pet food an arbitrary standard, and that the company says it notifies affected customers directly instead of issuing public recalls (dvm360, October 29, 2026). We covered the federal injunction case itself in our earlier Darwin's post; this post is about the two new cat food lots, and about how they were found.

An honesty note: the advisory does not say where in production the Salmonella entered, and it does not publish how much of the two lots was distributed or later accounted for. The later recall listing described the total quantity as unknown. Nothing below depends on the entry point — it depends on the detection point, which the record states plainly: a sick pet, reported by its owner.

The validation gap: a sick pet was the detection system

A raw pet food has no kill step waiting downstream. The cat eats it as it leaves the freezer, and freezing does not kill Salmonella. For a product built that way, the manufacturer's own testing, holding, coding, and recall systems are not supporting controls — they are the whole safety case. In this record, each one left the finding to someone else:

• Test and hold before the product ships, not after a pet is hospitalized. A finished-product pathogen program for a no-kill-step food needs a defined sampling plan, validated methods, and a genuine hold: lots do not move until results are in. When the first positive arrives via a regulator testing a complainant's food, the plant has effectively outsourced its release testing to its customers' veterinarians.
• Treat two lots, months apart, as a scope question — immediately. The two positive lots were manufactured on April 2 and July 8. A finding that spans production dates that far apart cannot be closed as a single-day event without evidence. The investigation has to ask what the lots share — ingredients, suppliers, equipment, environment — and hold or account for production in between until that question is answered.
• Validate the lot code for the product's real life, not its shipping life. This food can sit in a home freezer indefinitely, with no expiration date to retire it, and its identity sits on one of four units. Once that unit is used or discarded, the remaining units are anonymous — which is why FDA's fallback is simply to throw unidentifiable food away. Date coding, lot placement on every unit, and a defined shelf life are traceability controls, and they need the same validation attention as the process itself.
• Make recall effectiveness measurable. A subscription seller holds a record most manufacturers would envy: it knows exactly which customers received which lots. That advantage only counts if it is used and can be shown — quantities distributed, notified, returned, destroyed, and still unaccounted for, reconciled lot by lot. An email-only notification with the total quantity publicly unknown cannot demonstrate effectiveness to a regulator, whatever it achieved privately. As in the Viva Raw case, an action without evidence is a claim, not a control.
• Trend complaints as an early-warning system with teeth. Complaints will still arrive — no testing program samples everything. The discipline is in what a complaint triggers: preserved samples, retained-lot testing, a documented escalation threshold, and a hold first while the facts are checked. A complaint that has to travel to FDA and back before it produces a test result is a signal that spent weeks doing nothing.

None of these controls is exotic. For pet food manufacturers working without a kill step, the principle is simple: if your own system cannot find the problem, bound it, and account for it, your customers will find it for you — one animal at a time.

How a GMP Readiness Assessment catches it

This is exactly the failure pattern a GMP Readiness Assessment is built to find. Over four weeks, it reviews:

• Finished-product testing and hold-and-release — is there a sampling plan with real holding discipline for no-kill-step products, or can lots ship ahead of results?
• Complaint trending and escalation — what does a single illness complaint trigger, how fast, and who can place a hold on that evidence alone?
• Lot and date coding — does every consumer unit carry its identity, is shelf life defined for frozen storage, and can a customer still identify the product months later?
• Mock-recall reconciliation — can quantities distributed, notified, recovered, and outstanding be reconciled for one lot, from records, within hours?
• Investigation scoping — when positive lots sit months apart, is there a documented method for deciding what else is in scope?

It closes with a gap assessment, a remediation roadmap, and starter IQ/OQ/PQ templates — so the fix starts before the lab report does.