Checked this week: 9 entries, 8 axes. Most: ③ Tools & Agents, 2 entries. Fewer than the past 4-week average of 33 (4 weeks: 71, 20, 19, 23). 4 new actors — OpenAI, Oracle, Hemant Asirvatham, Elliot Mozeika, Parameter Labs.
9records added
8axes that moved
Nov 2–8dates
1,321total records then
Added, by axis
Axes with the most records come first. A record that spans several axes is counted once, under its main axis.
③2 records
Tools & Agents
Oct 28, 2026GitHub — GitHub published the Octoverse 2026 report — in the year to Sep 30, 2026 there were 17.9 billion pushes, more than 2024 and 2025 combined, while 34.3 billion commits and 1.3 billion merged pull requests were more than three times the figure of two years earlier, about 3.5 million a day. As of September 2026, pull requests opened by agents made up more than 40% of the total (17% in March) at 110 million, about six times the March figure, and 90% of them were merged within two weeks (83% in February). Among repositories created this year, those with an agent configuration file averaged 8 times as many commits as those without, and since June about 55% of all commits came from newly created repositories. There are more than 225 million developers, with 38.3 million in the United States, 32 million in India and 12.4 million in China. SourceThese are numbers the operator counted itself, showing that the activity figures of a code hosting service used by people have tilted toward agents. Pull requests opened by agents rising from 17% to 40% in half a year, with 90% of them merged within two weeks, means the automatically generated work is going into the main line rather than staying an experiment. By language, AI involvement was detected in about 40-44% of merged pull requests in TypeScript, Python and Rust, against 21-34% for Java, C# and C++, so ecosystems are taking it up at different speeds. The counting basis and the method behind AI involvement detection, though, are the company's own definitions and there is no outside reproduction.
Oct 27, 2026OpenAI, Oracle — OpenAI announced an expanded partnership with Oracle that brings Oracle apps and data into ChatGPT — Oracle AI Workspace, built by Oracle, lets users ask about information held in finance, procurement and planning apps in one place and finish approved business processes, and it arrives together with Oracle AI Security Mesh, which uses OpenAI cyber models to find and fix vulnerabilities, Oracle AI SMB Bundle for small and medium businesses, and Oracle App Composer, which builds apps with Codex. Oracle customers in the United States can buy the OpenAI API, Codex and ChatGPT Work through Oracle Marketplace using their existing purchasing process and Oracle Universal Credits. SourceOracle said that by its own count it has 130,000 active ChatGPT users and more than 95,000 active Codex users, and that preparing market research for its recruiting team fell from 2-4 days to 15-20 minutes. It means that the companies making core business software have moved past the stage of buying a model and bolting it on, and have begun moving their own apps inside the chat window. The shared design point is that the systems follow a company's existing permission and approval structures as they stand, and taken together with the ServiceNow expansion in the same week, entry routes on the business software side opened in several places within a month.
⑥1 record
Reasoning Models & the AGI Debate
Oct 26, 2026OpenAI — OpenAI announced that it would hand a set of mathematics, computational complexity and physics results produced by its internal models to outside research groups to be checked — it said it had proved Sogge's 1991 local smoothing conjecture in all dimensions, settling the Kakeya conjecture and the Stein restriction conjecture along with it (result Oct 18, 2026; formalized in Lean the next day); that it had shown the Exponential Time Hypothesis (ETH) to be false (result Oct 16, 2026; formalized in Lean the next day); and that it had proved the existence of four-dimensional interacting quantum field theories including gauge fields and fermions, which covers the existence of quantum chromodynamics and of pure Yang-Mills theory (Yang-Mills generated Sep 9, 2026, quantum chromodynamics Sep 18, 2026; formalized Oct 16, 2026). The complexity group is led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams and the physics group by David Gross, Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov; participants are not asked to sign non-disclosure agreements and are not paid. SourceIt put results produced by AI out as research raw material rather than as papers, and tested a way of releasing them that leaves the work of understanding and checking them to academics outside the company. It said it had applied a different release strategy to each of the three results — for the mathematics, only the material released; for the computational complexity, release and a research group in parallel; for the physics, the decision on release delegated to the research group — to see which is useful. The claim that the Exponential Time Hypothesis is false unsettles the premise on which hundreds of conditional hardness results rest, and the existence of Yang-Mills theory is part of one of the Clay Mathematics Institute's Millennium Prize Problems. All of it, though, is company-reported and independent review has not begun — a Lean formalization is a machine check of a proof's logic, not a guarantee that the problem was set up correctly. It gave $5 million to the Simons Foundation to cover the cost of review workshops and conferences.
⑨1 record
Open-Weight Lineage
Oct 29, 2026Mistral AI — Mistral AI released the weights of Mistral Large 4 (Le Chonk) — a 1 trillion parameter multimodal model with three more weeks of reinforcement learning added after the public preview. Company-reported, it leads Cognition's private FrontierCode 2.0 at 19.7% and rose on Terminal-Bench 4.0 from 28% in the preview to 41% while cutting output tokens by 30%. Its Artificial Analysis Coding Agent Index score of 48% is effectively tied with GLM-5.3, the top open-weight model, while using on average 47% fewer tokens per task, and its Cyber Index score is 51%. It offers reasoning in three modes, low, medium and high, and the company said its refusal rate for malicious cyber requests, at an average 96.6%, is the highest among the open models measured. It said the model was trained from scratch on European facilities. SourceA 1 trillion parameter model has appeared on the open-weight side putting coding and defensive security forward together, and the focus of the company's figures is not the score but how few tokens it takes to reach the same score. That is aimed at the situation where cost, not performance, becomes the limit first when an agent is left running for a long time. Putting a number on the line between doing defense and refusing attack requests, in the form of a refusal rate, is also rare for an open model. All the figures are company-reported, there is no independent reproduction outside the Artificial Analysis index, and FrontierCode 2.0 is a private evaluation, so outside checking is hard.
⑩1 record
Safety, Alignment & Interpretability
Oct 30, 2026OpenAI — OpenAI announced that it would put its community safety report on violence prevention on a regular schedule and tighten restrictions on weapons information for minors — it said it had passed more than 200 imminent threats of violence to law enforcement in the third quarter of 2026, and that the first report, covering the second half of 2026, would come in early 2027 and be updated every six months after that. Reconstructions of past mass attacks, the weapons and equipment used, and comparisons of the scale of casualties are restricted without asking age or reason; where a recent conversation shows signs of violence the model answers more conservatively, and the same restrictions apply when a user is estimated to be under 18. Conversations that fixate on violence are ended automatically, and repeated instances can lead to a temporary account suspension. It changed its age estimation system so that it is assessed on an ongoing basis, and adults who are misclassified can have that reversed through Persona, its age-check partner. SourceIt is a change that moves safety policy from what the model will not say to when a conversation is cut off and when an account is halted. It formalized the premise that a request that looks ordinary on its own reads differently when seen alongside earlier conversations, and said it adjusts answers on the basis of conversation history. For minors it shifted the standard toward blocking practical information about weapons and security devices even where no violent intent is visible. The promise of a report every half year creates material for outsiders to weigh the effect, but with the first report due in early 2027 there is for now only the company's own announcement. The count of referrals to law enforcement is company-reported as well.
⑮1 record
Regulation, Law & Governance
Oct 30, 2026U.S. District Court for the Southern District of New York — Motions for leave to file third-party amicus briefs arguing fair use arrived all at once in the consolidated OpenAI copyright litigation (1:25-md-03143) — on Oct 30 alone, separate motions were docketed from Midjourney and Runway, from Chamber of Progress and the Abundance Institute, from 10 AI startups supporting the plaintiffs (Splice, Boomy, ProrataAI, Created by Humans and others), and from former members of the U.S. Senate and House of Representatives. Judge Sidney H. Stein the same day denied one motion that had not followed the procedure for filing an amicus brief, saying it could be filed again. Docket entries reached number 2,463 by Oct 31, and no ruling on fair use has been issued yet. SourceThe docket itself shows the signal that a fair use ruling in this case is near. What stands out is that the line is not drawn between AI and creators — Midjourney and Runway, on the image and video generation side, are on one side, while AI startups whose business model distributes fees to creators have lined up with the plaintiffs. Taken together with OpenAI separately releasing a collection of briefs supporting fair use in the same week, the campaign for public opinion outside the courtroom and the briefs inside it are running on the same dates. No ruling has come yet, so this cannot be used as grounds for predicting the outcome.
⑳1 record
People & Discourse
Oct 28, 2026Hemant Asirvatham, Elliot Mozeika — Two OpenAI employees published The Future of the Firm, an essay that divides the org chart of a company mixing people and AI into three shapes — a promotion, in which AI sits beneath everyone and everyone becomes a manager; beside us, in which AI comes in as a colleague under a human manager; and between us, in which AI coordinates work between executives and staff — and wrote that all three will happen in part. It deals with the economics of the transition rather than full automation, and argues that the more capable AI is, the easier it is to adopt. It is the second piece in a series on the next-generation economy, and the authors stated that it does not represent the views of OpenAI or of their colleagues. SourceIf the workplace usage data and the enterprise software partnership OpenAI put out the same week are the story on the side of numbers and products, this piece is the story on the side of what that will look like inside an organization. The part worth noting is automation without layoffs — a picture in which you hand your former job to AI and move up a rung in the hierarchy, and in which, as the authors themselves admit, the price is the disappearance of the path where a new hire learned by doing the grunt work. The authors work at OpenAI and stated that this is not the company's view, so it should be read as a personal argument.
㉒1 record
Agent Security & Sandboxing
Oct 28, 2026Parameter Labs — Parameter Labs disclosed a chain of vulnerabilities that made it possible to take over another user's account through a single ChatGPT share link — the virtual keyboard listener in MathLive, the math typesetting library loaded by a graph placed in a conversation, did not check where a message came from, so a message sent by the sandboxed preview iframe could plant HTML in the parent page. That HTML created an iframe with no sandbox attribute and, riding a YouTube oEmbed callback that the Content Security Policy allowed, ran script with chatgpt.com privileges, read the session token and the account id, and sent out private conversations the attacker had no access to. It was checked on Oct 13, reported on Oct 18, and reproduced by OpenAI on Oct 21, which raised the severity from P2 to P1; it was closed on Oct 26 with fixes to both web frontends, and the bounty was $5,000. SourceIt is a case that follows all the way through how a preview, which shows what an agent made right where it was made, becomes a new attack surface. The sandbox was not broken; the ladder was the point at which the parent page trusted a message sent out from inside the sandbox, together with an external domain the Content Security Policy left open. All the victim did was open a share link, and the researchers said they went on to make the preview open by itself so that not even a click was needed. What divides the risk is less how you cage a model than where you render the model's output.
㉓1 record
On-Premises & Sovereign AI
Oct 30, 2026Cohere — Cohere released progress on a plan to put C$4 billion into research and development and AI facilities in Canada from 2024 to 2031 and to create more than 1,500 jobs — it said about 1,000 of those are technical roles in AI research, engineering and product development, and that it would create more than 250 paid student positions by 2029. It said new partnerships with the University of Toronto and the University of Waterloo, added to its existing relationships with Amii, Vector and Mila, form a network of universities and research institutes, and that through Bell AI Fabric and a C$309 million GPU cloud partnership it runs its own models on facilities inside Canada. SourceIt is a case of sovereign AI beginning to be spoken of in the language of industrial policy, bundling people, facilities and commercialization rather than model sovereignty. Cohere explained that it is spending on closing a gap, pointing out that Canada's national AI strategy targets a 3% increase in gross domestic product and up to 250,000 AI-related jobs by 2031, while only 12% of Canadian businesses said in 2025 that they use AI to produce goods and services. The amounts and the job counts are the company's own plan through 2031, and it did not separately state how much has been spent so far.
How this relates to the main moduleThis page holds only what was added that week. For the whole story, open the main module — it is one latest edition, revised every week, while weekly editions stay as they were.