# AI news, November 7, 2026: Qwen teaser, agent memory at Anthropic and BioBasicsBench _Saturday November 7, 2026 · AI provenance: This briefing was compiled by Claude (AI, Anthropic) from X search briefings by Grok Bot (AI, xAI), supplemented with primary sources. AI wrote the text; every claim has a source link. Responsible for llmnet.nl: Ivo Donker._ **Short answer:** The main story of November 6: Alibaba Qwen teases a new release without a name or date, Anthropic describes how Claude organizes its memory ('dreaming') and OpenAI introduces BioBasicsBench, a test for agents in laboratory work. DeepMind reports weaknesses in older 1024-bit RSA systems. Below, each topic comes with what it means and where it comes from. ## Alibaba Qwen teases a new release: 'The becoming begins' Alibaba's Qwen team posted a teaser poster on November 6 with the text 'The becoming begins'. The post drew about 15,500 likes. It names no model and no release date. QwenDevs amplified the teaser with its own post, 'Qwhen? Qsoon!'. What this means: something from Qwen is probably coming, but what and when is unknown. Anyone building on open-weight models does better to wait for the official release than to act on speculation. Sources: [@Alibaba_Qwen op X (teaser)](https://x.com/Alibaba_Qwen/status/2118674834632978755) · [@QwenDevs op X](https://x.com/QwenDevs/status/2118716745372532865) ## Anthropic describes 'dreaming': organizing and correcting agent memory Anthropic published an Engineering Blog on memory for agents. The company calls the process 'dreaming': Claude organizes what it has learned, corrects mistakes and can retrieve information faster. The announcement on X drew about 7,100 likes. What this means: memory is a core part of agents that handle longer tasks. How well this works in practice will only show in user tests; the blog itself is the source of the description. Sources: [@AnthropicAI op X (Engineering Blog)](https://x.com/AnthropicAI/status/2118814753154064501) ## Claude Code in the cloud handles dependencies and API keys; GitHub tests storage for 5,000 agents ClaudeDevs reports that in cloud sessions for Pro, Max and Ultra subscribers, Claude Code works out a repo's dependencies, network access and required API keys itself and saves the environment. The keys are stored without Claude seeing them. The post drew about 5,600 likes. GitHub says that in internal tests of a new Git storage architecture it reached 5,000 concurrent agents on one repository, with 7,200 pushes per second and low read latency. These are GitHub's figures from a test, not from production. What this means: vendors are building the infrastructure to let many agents work on code at once. Background on the tooling around this is in [our overview of MCP servers](https://apps.llmnet.nl/en/nieuws-mcp-agent-tooling). Sources: [@ClaudeDevs op X (cloud-omgeving)](https://x.com/ClaudeDevs/status/2118795192090386454) · [@testingcatalog op X](https://x.com/testingcatalog/status/2118844509346005399) · [@github op X (Git-opslag)](https://x.com/github/status/2118830274431873379) ## OpenAI measures agents' lab work with BioBasicsBench; Anthropic shows AI writing scientific models OpenAI introduces BioBasicsBench, together with biotech company Ligo. The benchmark consists of 100 tasks from real experiments and tests whether agents can interpret raw lab data, spot errors and know when the evidence falls short. The post drew about 2,800 likes. A guest post on Anthropic's Science Blog covers AI that writes, fits and revises scientific models as code, so that competing explanations for data can be explored side by side (about 2,400 likes). What this means: the focus is moving from chat questions to measurable tasks in laboratories. Scores from such a benchmark say something about the tasks in the test and not automatically about work in a real lab. Sources: [@OpenAI op X (BioBasicsBench)](https://x.com/OpenAI/status/2118782708767965238) · [@AnthropicAI op X (Science Blog)](https://x.com/AnthropicAI/status/2118792930211107115) ## DeepMind: AI agents find weaknesses in legacy 1024-bit RSA; OpenAI and the Frontier Model Forum publish a framework for cyber jailbreaks Google DeepMind shares research in which human-directed AI agents exposed two major weaknesses in older RSA systems with 1024-bit keys (about 1,000 likes). The post gives no details; anyone using this in a security assessment should read the research itself. Systems with 1024-bit keys have long been considered outdated. OpenAI and the Frontier Model Forum present a framework for judging when jailbreaks enable serious cyber harm (about 1,200 likes). A report from The Hacker News also circulates about an OpenSSH flaw (CVE-2026-123158, 'SSHole') that would bypass password authentication under certain Linux configurations and was reportedly exploited within hours. That report is secondary and unconfirmed here; administrators should check the official OpenSSH advisories. What this means: AI is becoming a tool both for defenders and for researchers hunting for weaknesses. Background is in [our overview of AI and cybersecurity](https://nieuws.llmnet.nl/en/ai-en-cybersecurity). Sources: [@GoogleDeepMind op X (RSA)](https://x.com/GoogleDeepMind/status/2118755418675765656) · [@OpenAI op X (kader jailbreaks)](https://x.com/OpenAI/status/2118795218099253606) ## Meta puts Muse Wonder models Active Memory and Active Focus live in Muse and Meta AI Meta reports that its specialized Muse Wonder models Active Memory and Active Focus now run in Muse and Meta AI. Meta itself cites a 2.4 times lower chance of forgetting critical information; the post drew about 570 likes. The measurement and the test setup are not in the post. What this means: Meta, too, is working on memory and attention in assistants. The figure is a claim by the company and not an independent measurement. Sources: [@AIatMeta op X](https://x.com/AIatMeta/status/2118768974779634084) Primary sources: [Alibaba Qwen op X](https://x.com/Alibaba_Qwen/status/2118674834632978755) · [Anthropic op X](https://x.com/AnthropicAI/status/2118814753154064501) · [ClaudeDevs op X](https://x.com/ClaudeDevs/status/2118795192090386454) · [OpenAI op X](https://x.com/OpenAI/status/2118782708767965238) · [Google DeepMind op X](https://x.com/GoogleDeepMind/status/2118755418675765656) [← Previous briefing](https://trending.llmnet.nl/en/briefing/2026-11-06) [Next briefing →](https://trending.llmnet.nl/en/briefing/2026-11-08)