Jamie Dimon warned about cockroaches in credit. The bigger threat is termites. Defaults are visible and episodic. Termites work in silence. Opaque AI-linked loans, cov-lite structures, and non-bank leverage are quietly hollowing out the beams of the bond market. The surface still looks sound: tight spreads, narrow volatility, and index-level calm. Underneath, we have mark-to-model assets, liquidity mismatches, and incentives that reward delay over discovery. When the load shifts, it will not be a headline default that fails, but the assumptions holding your portfolio together.
This year’s price action told you something changed. Leveraged loans, long treated as a floating-rate diversifier to high-yield bonds, have underperformed as spreads widened with the revaluation of AI-adjacent software and infrastructure borrowers. The usual positive correlation between loans and high-yield weakened right when investors expected it to hold. That is a red flag. It implies the risk engine many institutions rely on—assumed relationships between credit buckets—is calibrated to yesterday’s economy. AI hype pulled lending into companies with glossy growth decks, recurring revenue talk, and limited hard collateral. Now, as cash flows face higher rates and tighter budgets at customers, interest coverage weakens and equity cushions thin. Analysts note loan default rates north of 7 percent, with a build-up in distressed exchanges that avoid bankruptcy but do not fix solvency. Mohamed El-Erian has warned the AI bubble may not erase the system, but it can produce a chain of credit accidents. The signal is not a crash, it is the divergence itself.
Banks are stepping back where incentives conflict with price discovery. Some large lenders have reportedly restricted financing to private credit funds after marking down software loans. Non-banks filled the gap, extending highly leveraged loans at yields that failed to reflect true risk. A University of Bath study flagged underpricing among non-bank loans and the highest default rates since 2020, along with a rise in distressed exchanges. Why underprice? Because the model pays on volume, fee income, and reported stability. Monthly NAVs are smoothed. Hard bids are scarce until the gates creak. Interval funds, evergreen vehicles, and BDCs promised investors steady payouts from assets that do not trade daily. That is fine until redemptions test the model. Private-credit managers face a prisoner’s dilemma: sell early and realize losses, or hold and hope others blink first. When liquidity is model-based, not market-based, the first markdown is a policy choice. The second is forced.
Cockroaches are nasty but obvious—defaults that appear, get cleaned up, and clear price discovery. Termites hide in the covenant woodwork. They reside in cov-lite documentation, PIK toggles, and amend-and-extend deals that let borrowers avoid immediate failure while transferring risk down the pipe. From 2019 through the zero-rate years, lenders accepted erosion of covenants for a few extra basis points. Now that base rates are higher, the accumulated fragility surfaces. Companies survive with maturity push-outs and interest deferrals that disguise cash strain. Portfolios show low realized volatility, but accrual income masks capital decay. In engineering, microfractures propagate long before beams collapse. In credit, the analog is a rising share of borrowers needing special terms to keep paying. Defaults are a lagging indicator. The infestation is the rise of loans that can only pay if the window for refinancing remains open and spreads stay tight. That is not resilience. It is dependence on one state of the world.
Global regulators have flagged the problem: low risk premia, concentrated exposures to AI-linked growth stories, and leverage embedded in non-bank intermediaries. That mix is built for disorderly adjustments. Daily liquidity vehicles hold assets that settle on a good day and quote on a screen the rest of the time. ETFs give the illusion of exit; creation-redemption works until the underlying bids vanish, leaving discounts and a feedback loop into dealers, CLO warehouses, and margin calls. When spreads gap, redemptions and collateral tests arrive together. It becomes a coordination game that no one wins: sell now and take a hit, or wait and risk a bigger one. The first mover has the edge. That is why price impact is non-linear. The same 50 million that did not move the market last quarter can blow out levels when the street is short balance sheet. We saw versions of this in March 2020 in credit ETFs and again when UK pension funds met margin calls. A market built on smooth exit assumptions is anti-fragile only in PowerPoint.
CLOs were meant to turn rough loan risk into refined tranches. That works when correlations behave and recoveries track the handbook. Today, warehouses face mark-to-market pressures while equity tranches erode with each extension and markdown. Managers often hedge with broad high-yield indices. But when loans underperform high-yield because the pain is concentrated in AI-adjacent borrowers with loan-heavy capital structures, those hedges fail. The basis trade—long loans, short bonds—only works if spreads move together. Ask LTCM what happens when correlation assumptions go missing. Add in re-rating risk: if recoveries on covenant-light loans come in lower than modeled, mezz tranches that looked safe move. Basis hedges do not protect recovery shortfalls. The myth of contained risk rests on stable cross-asset relationships. This year’s divergence exposes that myth. Risk does not disappear in a tranche; it migrates and compounds when liquidity thins.
Many bond portfolios claim safety by citing low duration or high seniority. In a regime where rates stabilize or fall and growth cools, that logic backfires. Duration helps when Treasury yields decline, but if spread duration blows out at the same time, the net is not a cushion; it is a trap. Loans that float were sold as rate hedges. They also concentrate exposure to refinancing risk and earnings sensitivity to slower demand. The probabilities that used to be treated as independent—rate paths and credit outcomes—are now linked by AI-linked capital deepening that may not deliver cash returns on time. In plain terms: a single macro shock can hit multiple parts of the stack at once. History offers the base rate. After distressed exchanges, many firms default again when real cash needs return. Investors like to believe in mean reversion and diversification through labels. But this is a Minsky environment: stability hides the build-up of unstable structures. The math that ignores joint tails is not conservative, it is wrong.
Ignore the noise of quarterly beats propped up by cost cuts and capitalized expenses. Watch the micro data: the share of amend-and-extend deals, the growth in payment-in-kind interest, the creep in add-backs needed to meet covenants, and the depth of real bids in the loan market. Track the gap between model NAVs and executable prices in private credit funds. Look at who is funding who: when big banks step back from financing certain funds, they are not virtue signaling; they are reading collateral. Monitor the divergence across credit buckets for signs of broken hedges. And stop assuming AI as a theme guarantees borrower resilience. The Reserve Bank of Australia is right to warn that concentrated bets with low premia can unwind fast when leverage bridges the gap between hope and cash. Termites are not terrifying because they are big, but because by the time you see them, the wood is gone. The smart move in credit is not to spray after the swarm. It is to stop building with soft timber.
