> daily_signal(2026_07_13)

S&P just cut Oracle's credit rating to one notch above junk and named OpenAI the reason — roughly half of Oracle's $638 billion backlog rides on a customer that has never turned a profit.

PickBits Daily Signal · Monday, July 13, 2026

1. S&P just cut Oracle's credit rating to one notch above junk and named OpenAI the reason — roughly half of Oracle's $638 billion backlog rides on a customer that has never turned a profit.

For two years the question 'what happens if the AI build-out doesn't pay off?' has been a talking point. S&P Global just answered it in the only language that binds: a credit rating. It cut Oracle's long-term issuer rating from BBB to BBB- — the lowest rung of investment grade, one downgrade above junk, and the first time Oracle has been this close to it — and, unusually, it named a specific customer as the reason. OpenAI accounts for roughly half of Oracle's $638 billion in remaining performance obligations, about $300 billion of it tied to Project Stargate. Oracle is signing decades-long data-center obligations to serve five-year promises from a company that has never earned a dollar of profit; if OpenAI cannot pay, Oracle is left holding leases it cannot exit. S&P projects Oracle's free operating cash flow deficit widens to roughly negative $42 billion in fiscal 2027, nearly double its earlier estimate, with capex heading toward $95 billion. And S&P drew the distinction that matters: unlike AWS, Google and Microsoft — who can absorb spare AI capacity with their own internal workloads and who have far deeper reserves — Oracle has no fallback tenant. This is the AI trade meeting a hard external constraint for the first time. Not a short-seller, not a skeptical columnist: the credit market, pricing counterparty risk on the biggest customer in AI. Why it lands on a US reader: Oracle is a component of the S&P 500 and sits, via index funds, inside a very large share of American 401(k)s and pensions. The honest caveat — and it cuts against the alarm — is that BBB- is still investment grade, Oracle's stock ROSE on the news, and a rating action is a judgment about risk, not a prediction of default.

Key fact: Check whether you already own this risk without knowing it. Oracle is an S&P 500 component, so any total-market or S&P 500 index fund in your 401(k) or IRA holds it — this is not a story about traders. The concrete step: look up your fund's top holdings (every provider publishes them; search your fund's ticker plus 'holdings'), and see your actual exposure to the AI-infrastructure complex — Oracle, Nvidia, Microsoft, Broadcom — as a share of the fund. The point is not to sell anything. It is that 'I don't invest in AI' is, for most Americans with a retirement account, factually untrue, and S&P just repriced part of that exposure.

the-decoder.com · ground.news · hngn.com · primary source

2. Apple sued OpenAI in federal court on July 10, alleging it stole hardware trade secrets "at every level" — naming OpenAI's hardware chief, a 24-year Apple veteran, as the man directing it.

Continuing 05-17 #4, and the arrow has reversed. In May we reported that OPENAI had hired an outside law firm to consider suing APPLE for breach of contract over the Siri-ChatGPT integration. OpenAI never filed. Apple did — first, and on entirely different grounds. On Friday, July 10, 2026, Apple filed suit against OpenAI in the U.S. District Court for the Northern District of California, alleging a coordinated campaign to strip its hardware trade secrets. The two companies that put generative AI in a billion pockets are now at war in federal court. The complaint's language is unusually direct: 'at every level, from members of its Technical Staff to its Chief Hardware Officer, and in coordination with business partners, OpenAI has been stealing Apple's trade secrets and confidential information.' The named individuals matter. Tang Tan — OpenAI's Chief Hardware Officer — spent 24 years at Apple as VP of product design for iPhone and Apple Watch; Apple alleges he directed candidates interviewing at OpenAI to bring Apple secrets into the room. Chang Liu, eight years an Apple senior systems electrical engineer, is alleged to have kept an Apple laptop and downloaded confidential technical documents. Apple further claims OpenAI used its confidential project code names as recruiting bait, coached departing employees on evading Apple's exit-security procedures, and used a proprietary Apple metal-finishing technique without authorization. The backdrop is a partnership turned rivalry: ChatGPT was integrated into iOS in 2024; then OpenAI bought Jony Ive's design startup io for $6.4-6.5 billion and started building consumer hardware aimed squarely at the device Apple sells. OpenAI denies it: 'We have no interest in other companies' trade secrets.' What this actually decides, and why an IT-pro should care more than a gadget fan: trade-secret suits between rivals set the practical rules for ENGINEER MOBILITY. California voids non-competes, so trade-secret law is the lever employers use instead. A win for Apple on these facts makes 'what you carry in your head and your laptop' far more legally fraught every time a senior engineer changes AI employers. Note what Apple is NOT asking for: it seeks injunctive relief — barring use of the secrets, return of materials, evidence preservation — not, at this stage, a headline damages number. Ive is not named.

Key fact: If you are an engineer who may move between AI companies — which, in this market, is most of them — this case is about you, and there are three things to get right BEFORE you interview. (1) Never bring an artifact: no components, no slides, no code, no 'sample of my work' from a current employer. Apple's complaint specifically alleges candidates were asked to. Interviewers who ask are creating your liability, not theirs. (2) Return every device at exit and keep the confirmation. The complaint's most concrete individual allegation is an unreturned laptop with downloaded documents. (3) Understand that California voids non-competes, so trade-secret law is the tool your employer will use instead — and unlike a non-compete, it follows what you took, not where you went. Talk about your general skills; do not talk about your employer's unreleased roadmap.

techcrunch.com · axios.com · cnbc.com · fortune.com · washingtonpost.com · primary source

3. A Brown economics professor moved one exam from take-home to proctored, and his class average fell from 96% to 48.6% — the cleanest measurement yet of how much of an AI-era grade is the student.

Everyone in education has an opinion about how much students lean on AI. Roberto Serrano, who has taught Welfare Economics and Social Choice Theory at Brown for nearly two decades, accidentally ran the experiment. In spring 2026, for the first time, he gave a take-home midterm — students had asked for it, uneasy about sitting in a classroom after a mass shooting at Brown in December. The class average came back 96%. The historical average for that exam is 65-80%. Suspicious, Serrano and his graders fed the exam questions to ChatGPT and got back answers that mirrored what students had submitted — including the tell: rather than the obvious direct approach, many students had reproduced the same convoluted mathematical proof the model favored, 'kind of correct, but very off and with a very convoluted style.' He told the class he suspected widespread AI use and moved the FINAL to a proctored, in-person format. The average fell to 48.6% — the lowest he has ever recorded. The enrollment tells its own story: the class had swollen to 86 students from a typical ~30; after his warning, 18 dropped, 9 never showed for the final, and 19 failed outright. What makes this worth the slot is not outrage — it is measurement. The 47-point gap between the take-home and the proctored exam, in the same course, with the same students and the same professor, is the closest thing anyone has published to a controlled read on AI-inflated performance. It is also a US-immediate signal for anyone who HIRES: a transcript from an unproctored era encodes an unknown amount of model output. The honest limits, which Serrano concedes: he SUSPECTS but did not individually prove cheating; a proctored exam is harder than a take-home for reasons beyond AI (no notes, time pressure); and part of that 47-point drop is surely fear and attrition, not just absent AI. Brown's administration asked him to file individual complaints with exam copies — impractical at this scale — a response he calls 'meek' and 'appalling and insufficient.'

Key fact: If you hire, or you sit on an admissions or scholarship committee, treat unproctored coursework from 2023 onward as a weak signal and stop pricing it as a strong one. The concrete substitute is cheap and takes 20 minutes: give a short, live, proctored or synchronous task in the actual skill — a whiteboard derivation, a live code-read of a PR, a 'walk me through why you chose this approach' on work the candidate submitted. Serrano's 47-point gap is exactly the gap between what a take-home artifact claims and what the person can do unaided. You do not need an AI-detector (they do not work reliably); you need one unassisted observation.

insidehighered.com · the-decoder.com · thenextweb.com · primary source

4. OpenAI's o3 reread 376 children's rare-disease cases that specialists had already given up on — and 18 families, one waiting nearly two decades, finally got a diagnosis.

This is what the technology looks like pointed at a problem worth solving. Rare genetic disease is a diagnostic odyssey: a family spends years — often a decade or more — cycling through specialists, genomic panels and multidisciplinary reviews, and a large share simply never get an answer. Their cases get filed as unsolvable. Researchers at Boston Children's Hospital's Manton Center for Orphan Disease Research, with Harvard and OpenAI, took 376 of those closed, unsolved pediatric cases — cases that had ALREADY been through commercial and institutional genomic pipelines and expert teams — and reanalyzed them with OpenAI's o3 Deep Research model. It established 18 new diagnoses: a 4.8% additional diagnostic yield on cases the field had given up on. Ten came from 100 neurodevelopmental cases, four from 61 neuromuscular, two from 200 sudden-unexpected-death-in-pediatrics cases, and two from 15 early-psychosis cases. One of them is Kyra, now 28, who had waited nearly twenty years for the name of what she has: myofibrillar myopathy. The discipline here is the part worth defending, and it is why this belongs in the good column rather than the hype one. The model made ZERO clinical decisions. It produced evidence-linked candidate explanations for specialists to examine; a finding counted as a diagnosis only after expert review, additional testing, classification of the variant as pathogenic or likely pathogenic, confirmation by a CLIA-certified laboratory, and return of the result by the clinical team — with at least two team members reviewing each candidate and disagreements resolved by consensus. Before being pointed at the unsolved pile, o3 was tested on 51 already-SOLVED cases and recovered the correct gene and variant in duplicate runs for 48 of them. It worked in about six to ten minutes per case. The authors' own limits, stated plainly: this is retrospective, the number of new diagnoses is small, and the study did not measure time saved, cost, or whether care actually changed. It is not a product and it is not FDA-cleared. But 4.8% of 'unsolvable' is not a benchmark score. It is 18 children.

Key fact: If your family is living a diagnostic odyssey — a child (or adult) with a suspected genetic condition and years of inconclusive testing — the concrete, available-today step is REANALYSIS, and most families do not know to ask for it. Genomic data is not static: genes are newly linked to disease every month, so exome/genome data that was uninformative three years ago can be solvable now against current knowledge, and this study is a demonstration of exactly that. Ask your geneticist two questions: 'can my existing exome or genome data be REANALYZED against current gene-disease knowledge?' and 'am I eligible for the Undiagnosed Diseases Network?' (the NIH-funded UDN, undiagnosed.hms.harvard.edu, accepts applications from patients whose cases remain unsolved after standard workup). Reanalysis is often covered and does not require a new blood draw. Do NOT ask for o3 — it is a research protocol, not a clinical service.

techinformed.com · nbcnews.com · abcnews.com · openai.com · dataconomy.com · primary source