§ Claim under review · Fact
"The Fed is now quietly checking how much AI risk is hiding inside America's biggest banks. The New York Fed has been going into JPMorgan, Wells Fargo, Barclays and Morgan Stanley to look at their loans to private credit firms, according to Semafor. They're asking how big the exposure is and how good the collateral behind those loans really is... JPMorgan has already marked down big chunks of its loans that are tied to those software borrowers... private credit is paying for a huge part of the AI buildout, and many of those data centers won't make money for years."
Verdict
Credibly reported but unconfirmed
Confidence
MediumSummary
A viral post says the Federal Reserve is quietly checking how much AI risk is hiding inside America's biggest banks, citing Semafor. The Semafor report is real and dated 5 October 2026, and it does say New York Fed officials visited JPMorgan, Wells Fargo, Barclays and Morgan Stanley since the spring with questions about their lending to private credit firms, covering exposure size, risk management and collateral quality. However, the report rests on unnamed sources, and the Fed and all four banks declined to comment, so nothing has been confirmed on the record. The post also leaves out two things Semafor said: that it is not unusual for the Fed to go into banks when risk headlines pile up, and that the review had already concluded at some banks including JPMorgan. The framing as an "AI risk" check is a stretch, since the reported subject is private credit lending exposure, with AI-exposed software loans named as one trigger. The post also describes JPMorgan as marking down its own loans, when the reporting says it cut the stated value of loans held as collateral inside its private credit clients' portfolios, a preemptive valuation move rather than recognised losses. Barclays is a UK-headquartered bank, not an American one. The post ends with marketing for an investment product, which is separate from the news claim and was not part of this assessment. General information only, not financial advice.
The readings
key figures from the evidenceMorgan Stanley projected equity-debt split for AI buildout
Outstanding private credit loans to AI-related companies
Why this verdict
Evidence
The Semafor article exists and says substantially what the post says it says. The Federal Reserve Bank of New York has been visiting big banks to review their loans to private credit firms, and Fed officials have gone into JPMorgan, Wells Fargo, Barclays, and Morgan Stanley since the spring with questions about overall exposure, risk-management, and collateral quality, people familiar with the matter said.
The Fed, JPMorgan, Barclays, Wells Fargo and Morgan Stanley declined to comment.
Two things the post leaves out appear in the same article. The review was prompted in part by JPMorgan's move in March to mark down large swaths of loans in private credit portfolios, particularly those to software companies threatened by AI, and the Fed has concluded its review of some of the banks, including JPM.
It is not uncommon for the Fed to go into banks to review their holdings, particularly when there are frequent headlines about potential risks, and the Fed also does regular on-site evaluations. A separate summary of the same scoop notes the visits are separate from the Fed's routine on-site examinations and ongoing monitoring of bank risks , so coverage differs on how routine these visits were.
The March markdown event is independently documented. JPMorgan marked down the value of loans to software companies in the portfolios of private credit lenders, the Financial Times reported on 10 March 2026, citing unnamed sources, with software companies seen as vulnerable to the effects of artificial intelligence. The mechanism matters: the bank's trading division reduced the value of loans, most of which were made to software firms, sitting within the financing portfolios of private credit clients, and this was a preemptive move driven by market valuations, not actual loan losses .
On the AI buildout financing leg, research supports a large private credit role, framed as a projection. Across the full investment, Morgan Stanley projects roughly a 60-40 split between equity and debt, and within the debt component private credit is expected to account for the majority, about $800 billion or roughly 70 percent. One trade report states there are already more than $200 billion of outstanding private credit loans to AI-related companies .
A Fed official has addressed bank exposure to these lenders on the record. In a May 2026 speech, Governor Bowman noted that since 2015 the bank share of corporate lending decreased from 48 percent to 29 percent in 2025, and that loans to private credit vehicles "generally appear to be well collateralized, which should help protect against bank losses" . US supervisors are not alone: the ECB doubled the number of banks covered by a probe into links with private credit, asking more than 20 banks for details on their exposures, up from a dozen in previous exercises, according to people familiar with the matter .
Findings
✓ What's accurate 7
- Semafor published this report on 5 October 2026, and the post's attribution to Semafor is correct.
- The four bank names, the "since the spring" timing, and the three subject areas (exposure size, risk management, collateral quality) all match the Semafor text.
- JPMorgan did take markdown action in March 2026 on loans to software companies seen as exposed to AI disruption, reported by the FT on 10 March 2026 and covered by CNBC the next day.
- The AI-linked software markdowns were, per Semafor's sources, part of what prompted the review.
- Banks do lend substantially to nonbank lenders, and a Fed governor addressed this channel publicly in May 2026, describing those loans as generally well collateralized as of that speech.
- Private credit is a significant financing source for AI data centre construction, and projections place it as the majority of the debt component of that buildout.
- Supervisory attention to private credit is not unique to the US: the ECB expanded a similar exercise to more than 20 banks as of June 2026.
≈ What's misleading 6
- The post opens by saying the Fed is checking "how much AI risk is hiding inside America's biggest banks." The reported review is about banks' lending to private credit firms generally, with exposure, risk controls and collateral quality as the questions asked. AI-exposed software loans were one trigger named by the sources, not the stated scope of the exercise. The gap converts a credit-exposure review into an AI-risk investigation.
- The post drops Semafor's own two caveats, that such visits are not uncommon when risk headlines accumulate, and that the Fed had already concluded its review at some banks including JPMorgan. Both qualifiers cut directly against the "quietly" and "right now" framing.
- "That's what the Fed is trying to figure out right now" presents the exercise as live and ongoing across all four banks, while the source says the review had already concluded at some of them as of 5 October 2026.
- "many of those data centers won't make money for years" is stated as established fact. It is a forward-looking assertion about future project economics. The research I retrieved documents financing structure and scale, not realised or projected profitability timelines for individual data centres.
- The post says JPMorgan "marked down big chunks of its loans that are tied to those software borrowers." The reporting describes something narrower and different in kind: JPMorgan reduced the stated value of loans sitting inside its private credit clients' portfolios, which it holds as collateral, thereby cutting how much those clients could borrow. CNBC's source described it as preemptive and driven by market valuations rather than actual loan losses. The post's wording implies JPMorgan recognised losses on its own loan book.
- The post groups Barclays, a UK-headquartered bank, under "America's biggest banks." Barclays has large US operations supervised by US regulators, but it is not an American bank.
? What's uncertain 5
- Whether the review happened as described rests entirely on unnamed sources at a single outlet. Neither the Fed nor any of the four banks confirmed or denied it, and all declined to comment.
- The precise scope, depth and findings of any such review are unknown. Supervisory findings are confidential and were not reported.
- Whether the review is genuinely distinct from routine supervision is unsettled even within the coverage: Semafor frames the visits as not uncommon, while at least one secondary summary describes them as separate from routine examinations.
- The size of bank lending to nonbank financial institutions is cited inconsistently across the coverage, and a 2026 reporting reclassification affects the comparison. The post does not cite a figure, so this does not bear on its accuracy directly.
- Whether AI disruption will in fact impair the software borrowers whose loans were marked down is an open question about the future, not a settled fact.
Sources
8 of 9 linked to recordsSemafor, "Exclusive / Private-credit worries spur Fed review," 5 Oct 2026
Governor Bowman speech PDF, Federal Reserve Board, 8 May 2026 (NDFI and private credit lending)
CNBC, "JPMorgan Chase reins in lending to private credit firms after marking down software loans," 11 Mar 2026
Bloomberg, "ECB Expands Probe on Private Credit to More Banks," 15 Jun 2026
Brookings Papers conference draft, "Financing the AI Buildout," Sept 2026
Columbia Business School working paper, Van Nieuwerburgh, draft 20 Mar 2026
PYMNTS and City AM write-ups of the FT's 10 Mar 2026 JPMorgan story
SEC IAPD firm record, AUTOPILOT (CRD 331749, SEC 801-130594), and Form CRS dated 4 Feb 2026
Benzinga, investingLive, Newsquawk, ChainCatcher, Rallies.ai, First Squawk