Wellington's Sewage-Plant Review Puts a Proposed Downgrade Before the Benchmark

Running each through the benchmark's standard rounding rules alone, with no new conduct factored in, produces 58.7 for both: enough to cross from the Established band into Functional. For Netflix, the in-window news is a lawsuit Florida's Attorney General filed on 9 September 2026, alleging the company tracked children's viewing data on its Kids profiles without required consent and used it for advertising. Netflix told NBC News the suit "lacks merit" and said it would defend itself in court. Because the claim is unproven, it does not move Netflix's score on its own. For Bridgetown, the in-window news is a house fire that destroyed four homes on 3 September 2026. The government's response — rehousing the displaced families within hours and starting a rebuilding assessment five days later — is a positive story, not a negative one. Moving Bridgetown's score down here would misread the evidence rather than reflect it. Neither crossing was filed. Both published scores, 60.9, stand unchanged, and both cases were referred to a joint review of every entity that shares this same starting value.

A February 2026 law did extend the program only to the end of 2026, as reported. But a second, separate law — the Continuing Appropriations and Extensions Act, 2027, signed 2 September 2026 — extended the same program to 2028. Lesotho Times reported the bill, HR 6500, was signed into law that day; the bill number is supported by that report and a linked reference, though it was not independently verified against the US Congress's own records. This correction does not change Lesotho's score. The US decision is not something Lesotho's government did, so it is not counted toward Lesotho's score under the benchmark's rule that only an entity's own actions are scored. What does count — Lesotho has 27 labour inspectors covering about 33,000 registered businesses, and gave textile workers a smaller minimum-wage increase (4 percent) than other sectors received (4.5 percent) — was already reflected in Lesotho's score. Lesotho's score is confirmed at 35.9 out of 100, unchanged.

To guard against bias, the review allowed only one score change: a single upward move that a prior review had already committed to in writing back in July, contingent on Anthropic taking concrete security steps. That trigger was met. On 1 September 2026, Anthropic described new safeguards following earlier AI safety incidents, including blocking outbound network traffic by default, a real-time system to stop unauthorized actions, and moving about 150 engineers onto security work. Separately, a class-action lawsuit filed 8 September 2026 alleges Anthropic's paid Claude Max plan delivered far less usage than the advertised "5x or 20x" multiples. That claim is unproven and was not counted for or against Anthropic's score. Anthropic's score is confirmed at 59.1 out of 100, remaining in the Functional band. Given the conflict of interest, this specific assessment is disclosed openly so it can be independently verified.

It is the third round of layoffs delivered this way in 2026, and it landed in the same quarter Oracle reported $19.3 billion in revenue and $4.76 billion in net income. Because this is a repeat of conduct already reflected in Oracle's score, and it now recurs against a strengthening balance sheet rather than a struggling one, the benchmark records it as a further step down on the category that measures whether a company's internal treatment of people matches its external statements. Reported severance terms are four weeks of pay plus one week per year of service; whether this round again requires employees to sign away legal claims to receive severance, as an earlier round did, is not yet confirmed. Oracle's score is confirmed at 14.7 out of 100, in the benchmark's lowest band.

The settlement resolves the allegations without any determination of legal liability. Measuring Abbott's score against this new evidence produces exactly the same result as a downgrade already proposed on 26 August — from 57.8 to 50.0 out of 100 — so no second, duplicate finding was filed. The original proposal remains under review, now with independent support from a separate measurement three weeks later.

Tonight, new deportation rules that took effect 1 September 2026 pulled three separate categories down a step each. Measured on the standard formula alone, Sweden's number falls from 89.9 to 86.1 — a real decline. But because the published score is already held well below the formula's number, that decline shows up as a numeric increase of 4.8 points against the published figure. It is an artifact of comparing an adjusted number to an unadjusted one, not a real improvement, and Sweden's published score remains 81.3, unchanged.