Three flavours. Six best-before dates spread across five weeks of production. That is the scope of the recall Arla Foods issued on 23 October 2026 for its 200g Protein Pudding range, after glass fragments were found in the product. When a recall covers that many separate production runs rather than one suspect batch, the question for anyone running a dairy site stops being "how did glass get in" and becomes "why did nobody catch it for weeks." That second question is a glass contamination dairy product recall problem, not a one-off accident, and it sits squarely inside a prerequisite programme that most sites assume is working until an incident like this proves otherwise.
What Happened
Arla Foods Ltd recalled various Arla Protein Pudding products in Chocolate, Salted Caramel and Vanilla Cookie flavours, 200g packs, because they may contain glass. The Food Standards Agency published the alert as FSA-PRIN-51-2026 on 23 October 2026. Affected stock spans best-before dates of 11, 18 and 20 January and 6 February 2027 for Chocolate; 21, 23 and 31 January and 2 February 2027 for Salted Caramel; and 25 January 2027 for Vanilla Cookie. Point-of-sale notices went up in stores carrying the product, and the FSA's standard advice applied: do not eat it, return it to the retailer for a full refund.
Source: Food Standards Agency, FSA-PRIN-51-2026
No injuries had been reported publicly at the time of the alert. The FSA notice does not state where in the process the glass entered, and nothing in the public record identifies the source. What follows on that point is informed inference, not confirmed fact.
What Went Wrong
Glass is a physical hazard under Codex Principle 1, and physical hazards from glass are addressed through what BRCGS Issue 9 clause 4.7 calls glass, brittle plastic, ceramic and other brittle material controls: a site-wide register of every glass and brittle item in production areas, scheduled condition checks against that register, and a defined response when an item is found broken or missing.
The detail that matters here is the spread of best-before dates. A single cracked jar lid or one dropped light fitting typically shows up in one or two shifts of product, not six separate best-before dates running from mid-January into early February. That pattern is more consistent with a source that sat in place, undetected, across multiple production runs: a damaged glass viewing panel, a cracked sight glass on a vessel, or brittle glazing near the filling line that nobody flagged as broken until the contamination had already reached several weeks of stock. Finding glass in three different flavours also suggests the source sat upstream of where the flavours diverge, on shared equipment rather than on a single flavour-specific line.
The HACCP Perspective
Glass and brittle material control is a prerequisite programme, not a CCP. Dairy puddings are a poor candidate for in-line glass detection: metal detectors do not see glass at all, and X-ray systems struggle to distinguish small glass fragments from the natural density variation in a dairy matrix with fruit pieces, caramel or cookie inclusions. For most sites in this category, there is no validated CCP that reliably detects glass once it is in the product. That makes the PRP the entire control, not a backstop behind a CCP, and it is why a scheduled glass register with real verification matters more here than in categories where X-ray can be validated.
Codex Step 9 requires that control measures actually operate as designed, and Step 11 requires verification that the system is working, not just documented. A glass register that lists items but is only checked monthly, or where "all glass intact" gets signed off without anyone physically inspecting hard-to-reach fittings, satisfies the paperwork without satisfying the clause. The gap between a register that exists and a register that is verified is exactly where incidents like this one originate.
How This Could Have Been Prevented
A few checks worth running on your own site this week:
- Pull your current glass and brittle material register and confirm every item on it, light diffusers, viewing panels, sight glasses, gauge covers, was physically inspected in the last scheduled check, not just ticked off from memory.
- Check that your breakage procedure actually stops production and quarantines affected stock back to the last verified "all intact" check, not just back to the shift where the breakage was noticed.
- Confirm who is trained to spot a hairline crack in a sight glass or viewing panel, and whether that inspection is documented with enough specificity that an auditor can tell it happened versus was assumed.
- If your product category cannot support validated glass detection at the CCP level, make sure your risk assessment says so explicitly, rather than leaving glass control looking like it's backed by inspection equipment it is not.
Where SafetyCore Fits
SafetyCore's prerequisite programmes module links each PRP, including glass and brittle material control, directly to the hazard it mitigates in your risk assessment, so the connection between "this register" and "this hazard" is explicit rather than assumed. When a scheduled check is logged, it's timestamped and locked in the audit trail, and a missed or overdue check is visible rather than buried in a paper file nobody reviews until the annual audit. If an incident like this one happened on your site, SafetyCore's triggered review workflow captures the investigation and the corrective action against the specific Codex step it relates to, with evidence attached, which is what a BRC auditor or an FSA investigator will expect to see afterward.
Final Thought
A recall spanning six best-before dates across three flavours is not bad luck. It is what happens when a PRP that looks fine on a spreadsheet hasn't been physically verified in a while. Glass control is unglamorous, and it rarely gets the attention metal detection or allergen labelling gets in a risk assessment review. This recall is a reminder that for a lot of dairy and chilled categories, the glass register is not a backup control behind a detection system. It is the only control there is.