Nvidia Reports 103% Growth as OpenAI Brings GPT-6.2 to Copilot

OpenAI is positioning GPT-6.2 for both professional and software-development work. GPT-6.2 Astra targets demanding tasks, while Sol is intended for everyday use and complex agent workflows. Astra for Law combines the model with a legal search index and legal-writing instructions, and is available to ChatGPT Enterprise customers who request access through OpenAI’s Trusted Access Program, including through the API. On Harvey’s benchmark of 120 assignments across 24 areas of law, Astra for Law earned full credit on 20.8% of assignments, versus 15.0% for GPT-6.2 Astra and 13.1% for Muse Wonder Full. OpenAI also reported fewer material hallucinations than Claude Opus 5.5 in a comparison, and said the share of research answers passing every check rose from about 46% to 53.8% in less than two months. Some online users asked for EU certification.

The models also arrived in GitHub Copilot on November 18, with a gradual rollout to Pro, Max, Business and Enterprise subscribers across tools including Visual Studio Code, Visual Studio, Copilot CLI, JetBrains IDEs and Xcode. Astra focuses on advanced coding and terminal work; Sol supports agent workflows and code review. OpenAI says Astra improved on two coding benchmarks and its API token rates are 20% lower than GPT-6 Astra’s, while Sol keeps its predecessor’s pricing.

Nvidia’s results helped calm concerns that AI spending was slowing. The company beat quarterly expectations by 6%, reported revenue growth of 103% despite no China sales, and posted a 74.9% gross margin. Its shares rose almost 5% in after-hours trading after the November 17 report. Nvidia forecast margins of 71% to 72% next quarter and 72% to 73% in 2028. The report also lands amid uncertainty over leadership: interim CEO Collette Kress handled the investor call after Jensen Huang stepped down, and chairman Harvey Jones said the company is searching internally and externally for a CEO. Coverage says Amazon, Microsoft and Google remain committed to Nvidia’s AI solutions, while the company faces competition and congressional investigations. Executives and investors at Austin’s TomorrowX conference also defended Huang and played down market concerns, though the report provides few further details.

Research on shopping agents raises a reliability concern: across three product-choice benchmarks using WebShop, Amazon’s SQID and Wayfair’s WANDS, increasing the reasoning effort of ten leading language models did not meaningfully improve their choices. It did make them significantly more confident, even though most models were poorly calibrated and often made the same wrong selection. Separately, Adobe found AI-referred traffic to U.S. retail sites rose 71% year over year in September 2026, based on more than 1 trillion visits, 100 million products and 18 categories. The article cautions that brand visibility in AI answers alone does not show that the work generated customers or sales; retailers need evidence tied to their own results.

Several reports focus on AI’s risks and limits in customer-facing settings. Adults 60 and older filed more than 201,000 complaints with the FBI Internet Crime Complaint Center in 2025, reporting losses above $7.7 billion—about 37% more than the year before. Scammers use AI-cloned voices and convincing bank messages alongside government impersonation, romance and investment fraud; investment scams accounted for more than $3.5 billion in reported losses and cryptocurrency scams for $4.3 billion. The article advises hanging up and calling a trusted number, using a family code word, and avoiding gift cards or wire transfers. In a separate AnswerConnect survey of 6,000 adults in the U.S., U.K. and Canada, 55% said they were less likely to become customers if AI rather than a person answered, 59% found AI agents frustrating and 31% said they would hang up.

On enterprise AI infrastructure, Fortinet’s FortiAIGate is now a validated AI guardrail in Equinix’s Distributed AI Hub. It can inspect AI traffic, enforce policies and protect sensitive data across distributed workloads; Fortinet also offers FortiGate G series appliances designed for AI workloads. Another enterprise product, Resolve AI’s Primary On-call, is generally available to investigate production alerts using its 27-billion-parameter SRE-1 model. Resolve says the model solved about twice as many incidents as general-purpose coding agents in tests. Salesforce reported that the service closed more than 70% of CRM Analytics team paging alerts during its initial dark-launch testing; SRE-1 access remains limited to Resolve AI’s platform.

Two further reports offer limited detail. One says Anthropic’s planned IPO comes amid investor concern that AI enthusiasm has given way to fears of a bubble, but the provided text gives no further information about the offering. The Fortinet coverage notes a 29% net margin and a 42.4 price-to-earnings ratio, alongside a community investor’s estimated fair value of $125.56 based on assumed revenue growth and future margin figures; these are presented as estimates, not company guidance.

Key Takeaways

  • OpenAI’s GPT-6.2 Astra for Law earned full credit on 20.8% of Harvey LAB-AA assignments, compared with 15.0% for GPT-6.2 Astra and 13.1% for Muse Wonder Full.
  • GPT-6.2 Astra and Sol entered GitHub Copilot on November 18; OpenAI says Astra’s API token rates are 20% below GPT-6 Astra’s.
  • Nvidia reported 103% revenue growth, a 74.9% gross margin and results 6% above quarterly expectations, despite no sales from China.
  • Nvidia faces a CEO search after Jensen Huang stepped down; Amazon, Microsoft and Google are described as remaining committed to Nvidia AI solutions.
  • More reasoning effort did not significantly improve product selections by ten tested models, but it increased their confidence; test data included Amazon’s SQID and Wayfair’s WANDS.
  • Adobe found AI-referred traffic to U.S. retail sites rose 71% year over year in September 2026, but the article says visibility reports do not establish sales impact.
  • Older adults reported more than $7.7 billion in online-crime losses in 2025; AI-enabled voice and message impersonation is among the scam tactics described.
  • An AnswerConnect survey found 55% of respondents were less likely to become customers when AI answered instead of a person.
  • Fortinet’s FortiAIGate is available as a validated AI guardrail in Equinix’s Distributed AI Hub; Resolve AI’s Primary On-call is generally available and Salesforce reported it closed over 70% of specified test alerts.
  • The Anthropic IPO article describes investor concerns about an AI bubble but provides no further offering details.

OpenAI’s GPT-6.2 Astra targets legal work as users ask for EU certification

OpenAI introduced GPT-6.2 Astra for demanding work and GPT-6.2 Sol for everyday tasks, saying the models can handle more complex jobs while using fewer output tokens. Astra can help lawyers research cases, review contracts, draft court briefs and prepare legal documents. OpenAI’s Astra for Law configuration combines GPT-6.2 Astra with a legal search index and legal writing instructions. OpenAI said it ranked first on Harvey’s LAB-AA benchmark, which tests 120 assignments across 24 areas of law, and recorded 95% fewer material hallucinations per assignment than Claude Opus 5.5 in the comparison. Access is available to eligible ChatGPT Enterprise customers through the Trusted Access Program, while online reactions included requests for EU certification.

GPT-6.2 Astra and Sol bring stronger coding tools to GitHub Copilot

OpenAI’s GPT-6.2 Astra and GPT-6.2 Sol became available in GitHub Copilot on 18 November, with improved coding, reasoning and code review features. Astra focuses on advanced coding and terminal tasks, while Sol supports complex agent workflows and code reviews. OpenAI said Astra scored 5.2 percentage points above GPT-6 Astra on the Agents’ Last Exam benchmark and improved by 9 points on Terminal-Bench-Science 0.1. The Astra API has 20% lower input and output token rates than the GPT-6 Astra API, while Sol keeps its predecessor’s pricing. The rollout is gradual for Copilot Pro, Max, Business and Enterprise subscribers and includes apps such as Visual Studio Code, Visual Studio, Copilot CLI, JetBrains IDEs and Xcode.

OpenAI says Astra for Law leads legal AI tests

OpenAI released an updated Astra for Law configuration and said all ChatGPT Enterprise customers can request access through its Trusted Access Program, including through the API. The system pairs GPT-6.2 Astra with an expanded legal search index and instructions for legal research and writing. On Harvey’s LAB-AA benchmark, which tests 120 assignments across 24 areas of law, Astra for Law earned full credit on 20.8% of assignments, compared with 15.0% for GPT-6.2 Astra and 13.1% for Muse Wonder Full. It used about 64,000 output tokens per assignment, compared with 139,000 for Muse Wonder Full. OpenAI also reported that research answers passing every check rose from about 46% to 53.8% in less than two months.

Nvidia results ease investor fears over the AI boom

Nvidia reported strong results and forecasts on 17 November, helping ease concerns that spending on AI had slowed. Its shares rose almost 5% in after-hours trading. The company beat quarterly expectations by 6%, with revenue up 103% despite no sales from China, while gross margins were 74.9%. Nvidia forecast margins of 71% to 72% next quarter and 72% to 73% in 2028. Investors and analysts also noted uncertainty over leadership after Jensen Huang stepped down, with interim CEO Collette Kress handling the investor call and chairman Harvey Jones saying the company was conducting an internal and external CEO search.

Nvidia beats earnings expectations in its first report after Huang stepped down

Nvidia delivered strong earnings in its first major test without Jensen Huang, who stepped down amid allegations involving employee relationships and a brewing spy scandal. Interim CEO Collette Kress now faces a CEO search and questions about leadership, governance and geopolitical tensions. The article says the results suggest Nvidia’s customer relationships, product pipeline and market position remain strong, with Amazon, Microsoft and Google still committed to its AI solutions. However, the company also faces congressional investigations and rising competition. Investors welcomed the results, but the article says Nvidia must show it can maintain its AI chip position while restoring confidence and leadership stability.

AI Scams Put Retirees at Risk as Fraud Losses Climb

Elon Musk said even he could be fooled by highly persuasive artificial intelligence. In 2025, adults age 60 and older reported more than 201,000 complaints to the FBI Internet Crime Complaint Center, with losses topping $7.7 billion, about 37% higher than the year before. Scammers use AI to clone relatives’ voices and create convincing bank messages, while pushing old schemes such as government impersonation, romance fraud and investment scams. Investment scams caused more than $3.5 billion in reported losses, and cryptocurrency scams led to $4.3 billion in losses among older adults. The article advises people to hang up, call a trusted number, use a family code word and avoid gift cards or wire transfers.

More Reasoning Makes Shopping AI Agents Confident, Not More Accurate

Researchers tested ten leading language models on three product-choice benchmarks using data from WebShop, Amazon’s SQID and Wayfair’s WANDS. Increasing the models’ thinking effort did not improve the quality of their choices in a statistically meaningful way. However, it did significantly increase how confident the models said they were. Most models were poorly calibrated, meaning their stated confidence did not reliably match their accuracy. Models also often made the same mistakes and chose the same wrong product.

Anthropic’s Planned IPO Faces Investors’ Growing Doubts About AI

The article’s available text says stock prices began falling earlier this year for several major American companies. Its headline says Anthropic’s blockbuster initial public offering comes after a year when enthusiasm about AI turned into concern. The article may test whether investor fears about an AI bubble affect the company’s offering. The provided article text does not include further details.

AI Search Reports Need to Show Their Impact on Sales

A report showing that a brand appears in AI answers does not prove that the work brought customers or sales, the article says. Adobe found that AI-referred traffic to U.S. retail websites rose 71% year over year in September 2026. Its analysis covered more than 1 trillion visits to U.S. retail sites, 100 million products and 18 product categories. Retailers considering AI as a customer-acquisition channel need evidence tied to their own customers, services and sales.

Fortinet Adds FortiAIGate to Equinix’s Distributed AI Hub

Fortinet announced that its FortiAIGate security product is available as a validated AI guardrail in Equinix’s Distributed AI Hub. The product can inspect AI traffic, enforce policies and protect sensitive data across distributed AI workloads. The article says this move places Fortinet within Equinix’s AI infrastructure, alongside FortiGate G series appliances designed for AI workloads. It also notes that Fortinet’s near-term challenge is turning interest in AI security into deployments while protecting its 29% net margin. The stock trades at a 42.4 times price-to-earnings ratio, and the article cites a community investor’s estimated fair value of $125.56 based on 15.0% annual revenue growth and a future profit margin of 31.0%.

Survey finds phone callers prefer people over AI agents

AnswerConnect released an unedited recording of an AI sales agent that forgot why it had called and lost a potential customer. Its survey of 6,000 adults in the United States, United Kingdom, and Canada found that 55% were less likely to become customers when AI answered instead of a person. The survey also found that 59% find AI agents frustrating, 31% would hang up if connected to AI, and 34% say AI support rarely or never solves their problem. AnswerConnect CEO Natalie Ruiz said human receptionists listen, use judgment, and care about callers’ feelings. OnePoll conducted the survey in May 2026.

AI Summit leaders defend Jensen Huang amid market concerns

Fund managers, venture capitalists, and technology executives met at the TomorrowX conference in Austin, Texas, during the past week. They played down market concerns and defended Jensen Huang. The article says the AI market is experiencing repeated tremors, but the provided text gives no further details.

Resolve AI launches SRE-1 model for production alerts

Resolve AI has made its Primary On-call service generally available, using its SRE-1 model to investigate production alerts and find root causes. SRE-1 has 27 billion parameters and was trained by 16 AI researchers over 7.4 days, using up to 512 graphics processing units. In tests, the system solved about twice as many incidents as general-purpose coding agents, and an ensemble of five investigations matched the strongest tested frontier model at a lower cost, according to Resolve AI. Primary On-call can close alerts, mitigate problems when team policies allow, or escalate them to a person, and it checks whether fixes hold. Salesforce said the agent closed more than 70% of CRM Analytics team paging alerts during its first few weeks of dark launch testing; access to SRE-1 is limited to Resolve AI’s own platform.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

OpenAI GPT-6.2 GPT-6.2 Astra for Law AI coding assistants GitHub Copilot legal AI AI benchmarks AI hallucinations Nvidia earnings AI infrastructure AI shopping agents AI agent reliability AI-driven retail traffic AI scams voice cloning customer service AI Fortinet FortiAIGate Equinix Distributed AI Hub Resolve AI SRE automation AI investment bubble

Comments

Loading...