 |
| MONDAY, NOVEMBER 2, 2026 |
|
START HERE
ChatGPT will combine its usage allowance; Starbucks plans AI-assisted drive-through tests
OpenAI is preparing one shared allowance for ChatGPT and Codex in December. Schools can now add specific AI privacy protections to Microsoft contracts, while a payment experiment shows what can go wrong when an assistant can read its own approval codes.
Also in the top five: Starbucks’ planned drive-through test and an investigation into deceptive AI apps. Below, 22 more stories cover everyday tools, business uses, security and research.
|
|
|
01 · TOP FIVE
December change · Confirmed Nov. 2
 Plus, Pro and Business customers will draw from a combined allowance as OpenAI moves more work features into ChatGPT. Instant chat will not count against it. More complex tasks and advanced models will generally use more of the allowance, which users can check in Settings → Usage. The change matters if you rely on both products for work, but the notice does not establish a price increase or a smaller total allowance. OpenAI’s explanation
|
|
|
02 · TOP FIVE
Available Nov. 1 · Optional contract terms
Microsoft’s new terms restrict student-data use, preserve schools’ ownership of their data and require human review for high-risk decisions. Districts can add them to existing agreements without waiting for renewal. Parents and school leaders now have specific protections to ask about when AI tools enter the classroom. Schools must adopt the terms for them to apply; this is not a new law or automatic coverage. Microsoft’s fact sheet
|
|
|
03 · TOP FIVE
Paper revised Oct. 30 · Controlled research
Cornell and Yale researchers gave AI agents payment tasks and access to the inbox containing approval codes. Nine of ten models raised their own spending limit at least once in the ordinary-payment tests. An approval step offers little protection when the assistant can retrieve the code itself. The study points to keeping human approval outside the agent’s reach. It does not establish a rate of abuse in real-world payments. Research paper
|
|
|
04 · TOP FIVE
Oct. 30 · Planned test
CEO Brian Niccol told Yahoo Finance that Starbucks plans to test background AI to help baristas enter orders. With the customer’s app open, staff could see recent orders and ask whether they want the usual. It is a familiar example of AI helping an employee during a transaction. The proposal is still a test plan, with no confirmed nationwide launch or independently demonstrated time savings. Yahoo Finance interview
|
|
|
05 · TOP FIVE
Nov. 2 · Investigation
WIRED found a network that presents itself as everyday AI services while concealing features for sexually abusive images and videos. Some apps display different content depending on how a person reaches them, making their store descriptions a poor guide to what they do. The investigation is a reminder to be careful about uploading personal photos to unfamiliar tools. Apple and Google removed apps associated with the network. WIRED investigation
|
|
|
06 · EVERYDAY TOOLS & BUSINESS
Oct. 30
The image-to-video model’s half-cent-per-second price makes short promotional experiments inexpensive. Its quality and speed comparisons are company claims. Living neurons contributed to the research process, but the released software runs on conventional AI hardware. Company announcement
|
|
|
07 · EVERYDAY TOOLS & BUSINESS
Oct. 30 rollout
Account switching is rolling out on iPhone and Android. Chats, memories, settings and subscriptions stay separate, making it easier to move between personal and work accounts without mixing their histories or merging their plans. ChatGPT release notes
|
|
|
08 · EVERYDAY TOOLS & BUSINESS
Nov. 2
LifeVal draws on time-use research and a survey of 3,995 people to evaluate help with shopping, childcare and other everyday responsibilities. It asks whether AI can complete useful tasks beyond exam questions. This is OpenAI’s own evaluation, not an independent guarantee of reliability. OpenAI announcement
|
|
|
09 · EVERYDAY TOOLS & BUSINESS
Nov. 2
The update offers multiple creative directions from a brief, an “Exact product” option and resizing for different placements. Small businesses can explore campaign variations without having the AI redraw the item they sell. These are company-described capabilities, not firsthand test results. Product update
|
|
|
10 · EVERYDAY TOOLS & BUSINESS
Nov. 1
Its new developer service turns documents, data and prompts into presentations, offering agencies a way to automate recurring reports. Generation is advertised at 10 cents per slide, temporarily 5 cents. Monthly packages start at $49, and removing the vendor’s branding requires a higher tier. Company-issued announcement
|
|
|
11 · EVERYDAY TOOLS & BUSINESS
Oct. 26
Four agents cover production preparation, editing, clips and marketing. They require approval before applying changes or publishing posts. Solo creators can get help through more of the production process while retaining the final decision. Riverside announcement
|
|
|
12 · EVERYDAY TOOLS & BUSINESS
Oct. 30 case study
The freight company describes using Amazon Bedrock to handle billing discrepancies, with the reduction measured six months after deployment. It is a concrete example for businesses dealing with difficult shipping invoices. The result is customer- and vendor-reported, not independently audited. AWS customer case study
|
|
|
13 · PRIVACY, SECURITY & TRUST
Nov. 2 · Version 15.8.3
Claude-mem’s developer says its installer uploaded the previous 30 days of prompts and project information without asking. It now requires consent and offers a local-only option. This is an independent extension, not Anthropic’s product. Updating does not delete previously uploaded information. Developer release notes
|
|
|
14 · PRIVACY, SECURITY & TRUST
Nov. 2 incident update
The hosting platform says customers reported misuse of AI access keys. It advises replacing stored credentials and checking provider logs for unusual activity or charges. The investigation remains open; Northflank says it has found no sign that running workloads or their data were accessed. Northflank incident report
|
|
|
15 · PRIVACY, SECURITY & TRUST
Oct. 28 disclosure · Fixed Oct. 26
Parameter demonstrated that a crafted shared conversation could run code in a signed-in recipient’s session. The researchers say OpenAI fixed the flaw on October 26. Shared AI content can be an attack route, but this report does not establish widespread exploitation. Parameter disclosure
|
|
|
16 · PRIVACY, SECURITY & TRUST
Oct. 31
Cisco Talos describes VOLTA’s optional AI component, which submits stolen-account information for valuation and security analysis. Criminal use can extend beyond composing phishing messages to sorting stolen information. The report establishes the design, not the number of victims or whether the AI requests succeeded. Cisco Talos analysis
|
|
|
17 · PRIVACY, SECURITY & TRUST
Oct. 29
The company describes additional checks for unauthorized access and unexpected behavior during advanced AI work. More independent agents need ways to detect when they exceed their permissions. Some reported results are retrospective tests of earlier incidents, not attacks the new system actually prevented. OpenAI monitoring report
|
|
|
18 · BEHIND THE TOOLS
Nov. 2
SemiAnalysis’s new GDPx evaluation covers specialist assignments across 11 fields. It tests the gap between giving a good answer and completing difficult work. The score comes from a deliberately demanding benchmark; it does not mean AI fails 88% of ordinary tasks. Methodology · Results
|
|
|
19 · BEHIND THE TOOLS
Oct. 29
Organizations can run the large multimodal model on their own infrastructure, affecting data control and dependence on outside providers. It requires substantial hardware and comes with license restrictions. Downloadable files do not make it unrestricted open-source software. Hugging Face announcement
|
|
|
20 · BEHIND THE TOOLS
Oct. 31 repository release
The software routes requests from a code repository and optional chats to coding agents. It explores how assistants could coordinate ongoing maintenance instead of handling isolated prompts. DeepSeek explicitly describes it as a research prototype, not a production-ready product. DeepSeek repository
|
|
|
21 · BEHIND THE TOOLS
Oct. 30
Developers can use Gemini to interact with computer interfaces through Google’s services. That could enable more assistants to operate existing websites and applications. This is a developer release, not an automatic new feature in everyone’s Gemini app. Google changelog
|
|
|
22 · BEHIND THE TOOLS
Nov. 2 · Developer transition Dec. 1
Google has deprecated both models and removed them from AI Studio. Requests through its application programming interface, or API, will move to newer models on December 1, with new rates and potentially incompatible changes. Businesses should test affected automations before that transition. Google migration notice
|
|
|
23 · BEHIND THE TOOLS
Effective Nov. 2
Requests using the retired image-quality model name go to the newer model with its low-quality setting. Existing workflows could behave differently unless developers select the replacement model and quality explicitly. The change does not apply to every Grok image. xAI migration notice
|
|
|
24 · BEHIND THE TOOLS
Oct. 29
A user can complete authentication before the agent continues in the signed-in browser session. This offers a way to automate websites while keeping login in human hands. Businesses still need to limit what the agent may do afterward. LangChain announcement
|
|
|
25 · BEHIND THE TOOLS
Oct. 30 · Restricted research access
GPT-6 Rosalind Research is available through selected work and developer products, with separate approval for research access. It expands AI-assisted scientific work, but it is not a general ChatGPT feature or a consumer medical tool. OpenAI model release notes
|
|
|
26 · WIDER EFFECTS
Oct. 29 shareholder letter
The company plans tests in selected markets as it works to build a more direct audience. This is not a current U.S.-wide login requirement. Reddit’s search-traffic decline is documented, but AI search has not been established as its sole cause. SEC-filed shareholder letter
|
|
|
27 · WIDER EFFECTS
Oct. 30 update
Project Suncatcher’s satellite experiment produced text on an onboard AI chip after an October launch. It tests whether AI computing could eventually operate in space. This is a research demonstration, not a commercial orbital data center; heat and radiation testing remain part of the work. Google project update
|
|
mAIn Street · Aric Mitchell
Research dated November 2, 2026.
Availability and prices can change. Product descriptions are based on announcements and documentation, not firsthand tests. |