Credit Suisse entered March 2023 reporting a CET1 ratio of 14.1% and a liquidity coverage ratio near 150%. Both numbers were accurate. Both were published. Within a week, the bank had been merged into UBS over a weekend under Swiss emergency law.
The easy conclusion is that capital regulation failed. The easy conclusion is wrong, and getting it wrong matters, because the wrong diagnosis produces the wrong reforms.
Start with what did not happen. No going-concern ratio was breached at any point. There was no measured insolvency. On every metric the framework asks banks to publish, Credit Suisse was compliant to its last day. The bank died of a loss of confidence, and a solvency ratio is not a confidence measure. It never was.
CHF 15.8bn of Additional Tier 1 was written down to zero under an emergency ordinance, at the point of non-viability. The instruments did not reach their contractual CET1 trigger. Equity holders received CHF 3bn.
Read that sequence again, because it inverts the hierarchy every AT1 investor believed they had bought. The instruments designed to absorb losses ahead of equity absorbed everything, while equity absorbed less than nothing. The global AT1 market froze. EU and UK authorities issued statements the same week committing to respect equity-first loss absorption, which tells you how quickly the official sector understood the pricing problem it had just inherited.
The mechanism matters more than the outcome. The AT1 did not convert because a trigger was hit. It was extinguished at the point of non-viability, with reported CET1 sitting above every contractual trigger in the documentation. That is not a malfunction. Since January 2013, every AT1 and Tier 2 instrument has been required to absorb losses at the point of non-viability, either through a contractual term or through a statutory resolution regime achieving the same effect. Here the authorities invoked both routes in a single decree, the instruments’ viability-event clauses and the emergency ordinance, and the first-instance court has rejected both: the live question in the litigation. What is settled is that the ratio-based trigger was never reached, because it never is.
AT1 absorbed nothing while Credit Suisse was a going concern, and everything at the moment it stopped being one. The going-concern label on the asset class describes intent, not observed behavior. Mechanical triggers set at 5.125%, or even a high-trigger 7%, sit so far below the level at which a modern bank loses its funding that non-viability reliably arrives first. Credit Suisse did not disprove the design. It demonstrated it at scale.
This is not a closed case, and anyone writing about it as settled history is ahead of the record. The Swiss Federal Administrative Court declared FINMA’s write-down decree unlawful in a lead-case partial decision of 1 October 2025, on the basis of insufficient legal authority. FINMA and UBS appealed to the Federal Supreme Court, which granted suspensive effect, so the instruments remain written down pending final judgment. Roughly 360 domestic complaints sit suspended behind the lead case, with investment-treaty claims accumulating abroad.
The largest AT1 loss event in history may yet be unwound in damages. Hold that thought next time someone tells you bail-in is legally robust because the term sheet says so. Subordination is only as good as the jurisdiction that has to enforce it at five o’clock on a Sunday.
Read alongside SVB and First Republic, the cohort supports a narrower and more useful conclusion than either of the popular readings.
The post-2010 capital quantum was not the problem. No 2023 failure was a capital-adequacy failure in the measured sense, and the recapitalized system absorbed the shock without contagion to the well-capitalized core.
Four other things failed instead:
Measurement perimeter: the AOCI opt-out kept unrealized securities losses roughly the size of SVB's equity outside regulatory capital.
Instrument design: the going-concern fiction described above.
Coverage: going-concern requirements stopped at eight US banks.
Speed assumptions: buffers and MDA glide paths built for a quarterly world, facing outflows measured in hours.
Every one of those is a design seam visible in the framework’s own fine print long before 2023. That is the uncomfortable part. The framework was not surprised by these failures; its own text had flagged each of them as a compromise.
Switzerland is legislating, and its capital core is not about AT1 design at all. The Federal Council adopted its Banking Act dispatch on 22 April 2026, requiring full CET1 backing of foreign participations at the parent bank, up from roughly 45% today, over a seven-year transition. The official impact estimate is around USD 20bn in additional parent CET1 requirements for UBS, an effective shortfall of near USD 9bn. It has been before Parliament since summer 2026.
AT1 design reform, in the jurisdiction where the AT1 question was created, has been explicitly deferred pending international developments. The paper records no EU, UK or US position on the instrument’s design either.
That is the honest state of play three years on. The epicenter is legislating what the paper calls the largest single national capital recalibration it records as in train, measured by the additional requirement at one bank, and it targets balance-sheet structure and group perimeter. Meanwhile, the reform options for the instrument whose failure generated the headlines remain on the table, with none adopted.
Two habits follow from taking the episode seriously.
First, stop reading distance to MDA as distance to trouble. Credit Suisse reported capital ratios comfortably above requirements to its last day, and no going-concern ratio was breached at any point. The relevant distance in a confidence crisis is to the point at which an authority concludes the bank is no longer viable, and no ratio in your reporting pack measures it.
Distance to MDA is a static, quarterly number. Distance to a confidence crisis moves in hours. Closing that gap is exactly what stress testing frameworks like Mirai ALM & Liquidity are built to monitor continuously, rather than at the next reporting date.
Second, know which of your instruments absorb losses by contract and which by statute, and in whose courts. For most banks the honest answer is that the going-concern layer above CET1 is gone-concern capital wearing a going-concern label, priced by a market that has now watched the label fail once and is waiting to see whether a court agrees.
The capital quantum was never the issue at Credit Suisse. What the episode put on trial was the numerator’s design, and that trial is still running in a Swiss courtroom.
For the complete architecture behind these numbers, including the going-concern versus gone-concern distinction, TLAC design, and how the EU, UK and US have each responded to the 2023 failures, read the full guide, "Capital Regulation: A Complete Reference — From Zero to Expert: Basel, the EU, the UK and the US Frameworks."
To see how Mirai Regulatory Reporting keeps capital, buffer, and TLAC reporting aligned across Basel, EU, UK, and US requirements, without waiting for the next stress event to expose a gap, learn more here: https://mirairisktech.com/regulatory-reporting