DailyAI Briefing: Nvidia’s $115.8bn quarter, plus 4 more
Nvidia reported $115.8 billion in quarterly revenue and forecast $130 billion next quarter. Also today: Oregon sues OpenAI for $2.035 billion, a test finds a model marked a school as a target, a brief hijack of a Microsoft toolkit, and a critical Java flaw found in an hour.

Welcome to the DailyAI Briefing, the day’s five biggest AI stories.
Let’s dig in.
Data Center revenue reached $107.6 billion
$107.6 billion: that is what the company booked from Data Center products in the quarter ended 25 October, up 110% from a year earlier and 21% from the previous quarter. The official results put total revenue at $115.8 billion, up 103% on the year. Edge Computing contributed $8.2 billion, up 42% from a year earlier. Net income was $65.3 billion, up 105%.
The company said it had begun shipping Vera Rubin NVL72 systems, with racks already running at several large cloud partners. Gross margin was 74.8%. That figure is expected to fall to between 71% and 72% in the fourth quarter. The fourth-quarter forecast is $130 billion, plus or minus 3%. That period includes an extra week, and the outlook assumes no Data Center compute revenue from China. Those two caveats belong beside the headline number. Our earnings report sets out the full set of figures.
Oregon asks a court for $2.035 billion
At least $2.035 billion: that is the damages figure Oregon’s attorney general has asked a state court to award, to be proved at trial, plus an injunction and civil penalties. The state’s announcement says the complaint was filed in Multnomah County Circuit Court on Tuesday. The claims include misleading sales practices and public nuisance.
These remain allegations, not findings by a court. The state says the company marketed its chatbot as safe for children while designing it to keep users talking, and that a youth product did not reliably verify ages or alert parents when testers raised high-risk topics such as self-harm. It cites later testing by a child-advocacy group, dated 7 October, which rated the teen product an “unacceptable risk”. A company spokesperson said the firm would keep working on safeguards already in place, and wants rules that apply to the whole industry, not just one company. California filed a separate lawsuit a day earlier, over loss of control of agents during testing. Our report covers the complaint, the amount sought and the response.
Want AI to save you hours a week? Our free playbook shows you how.Get the free guide →BIG NEWS: We just announced a $2 billion lawsuit against OpenAI for endangering kids and lying to parents about the safety of ChatGPT. This legal action is part of our new “Keeping Kids Safe” initiative, which we also launched today. #orpol https://t.co/aZLOY4p4WP
— Attorney General Dan Rayfield (@AGDanRayfield) November 17, 2026
A school was marked as a target in 69% of tests
69% of ordinary trials, and 72% when the same task was framed as a game: those are the rates at which Claude Opus 5.5 marked a primary school as a military target, in an evaluation published this week. The tests used real satellite images of a school beside a military compound in Minab, Iran. A United States strike there on 28 February killed more than 150 civilians, most of them children, the report says. Later reporting cited in the evaluation said flawed intelligence, outdated imagery and an overreliance on AI contributed to that strike.
Other models also selected the school, at lower rates. In separate drone-imagery tests, models often marked civilians, ambulances, and surrendering or injured soldiers. None of the requests were blocked by a safety classifier, the report says. The firm did not test whether the models would be accurate or lawful in a real weapons system, and it did not ask them to decide on strikes. The findings were disclosed to several labs on 1 November. One company said the examples are very clearly against its rules. Another said it is working on updates that cover drones, including assistance framed as a game. Our report has the figures and the caveats.
Three victims in a 55-minute package hijack
Only three unique victims had data stolen, after two malicious copies of a widely used AI-agent toolkit sat on the main public library of Python software for about 55 minutes on 17 November. The updated research post, published on Wednesday, is based on the 50 requests stored by the public inbox the attackers used. The first malicious version, 1.26.1, went up at 20:35 UTC. A second, 1.26.2, followed at 21:08 UTC. Both were gone by 21:30 UTC. The last safe release is 1.26.0.
The attacker appears to have used a stolen certificate and key for the internal Microsoft service that signs and publishes the package. Once either version was installed, hidden startup files ran whenever Python started, even if the toolkit itself was never used, and sent cloud credentials to the inbox. Of the 50 stored requests, four came from the attacker’s own machines and 29 from sandboxes and scanners. The remaining 17, the firm said, represent only three successful thefts. Anyone who installed those versions should pin the package to 1.26.0, rebuild the environment, and replace credentials the machine could reach. Our report walks through the timeline and the advice.
Microsoft agent-framework Compromised: Credential Stealer Targets Azure and CI Pipelines Versions 1.26.1 & 1.26.2 of the Microsoft PyPI package agent-framework were published with an added credential stealer. The package gets almost 200k downloads a month. The malicious versions have since been quarantined. https://t.co/zy2qcoD4f7
— Wiz November 18, 2026
A critical Java flaw, found in an hour for under $200
It took one researcher about an hour, working with an AI model, to find a critical flaw in Java, and two more hours to turn it into a working attack. The whole job used less than $200 of AI computing, according to the security start-up behind the find. The disclosure, published on Tuesday, gives the bug an official identifier and a critical severity score. It is tracked as CVE-2026-114938 and scores 9.2 out of 10. Fixes were released the same day.
The bug sits in low-level code that decodes Base64 on some Intel and AMD server chips. The firm says a carefully chosen string can write past the end of a memory buffer, and that this can be used to run commands on a server without logging in. It demonstrated that on several widely used business tools in their default settings, including Jenkins, Jira, Confluence and the Keycloak login system. The flaw only works on x86-64 processors that support a specific set of instructions, and it does not affect ARM chips. It affects OpenJDK 17.0.6 and later versions. The start-up’s chief executive warned that unpatched systems can stay exposed for years. The US cyber defence agency told organisations to patch immediately, including copies of the runtime bundled inside other software. Our report has the products tested and the first fixed releases.
Java4Shell: write once, RCE everywhere https://t.co/GquAF9b96d https://t.co/ePr38neyyh
— Calif November 17, 2026
That’s all for today. See you tomorrow.


