AI Daily · Nov 3
Nov 2 06:00 – Nov 3 06:00 (Beijing time) · 18 items
Updated 11/3 08:09 (Beijing time)
Today's highlights
- AnthropicAnthropic Messages API launches message threading in public betaAnthropic has launched message threading in public beta on the Messages API. The server caches multi-turn conversations, so developers only need to send the new message and the previous message ID instead of resending the full conversation, allowing Claude to start responding faster.
- Related companies · AnthropicGPT-6.2 Astra enters limited beta testing; Claude Mythos 5.5 leaked the same dayOn Nov 2, multiple users posted on X that OpenAI appears to have begun testing the GPT-6.2 Astra model, which has already appeared in Codex usage statistics; the same day, Claude Mythos 5.5 was exposed, with weights allegedly leaked on HuggingFace and downloaded 5,000 times before being taken down.
- Related companies · xAIMusk responds to rapid Grok Bot quota consumption: costs too high to sustainGrok Bot users' weekly quotas have been lowered and are being consumed faster. Musk says the service's costs are too high to sustain, that lowering quotas is the rational move, and that the product has "the fastest growth rate of any product I've seen in my lifetime."
Models and products
AnthropicAnthropic Messages API launches message threading in public beta
Anthropic has launched message threading in public beta on the Messages API. The server caches multi-turn conversations, so developers only need to send the new message and the previous message ID instead of resending the full conversation, allowing Claude to start responding faster.
xAIThird-party reportMusk responds to rapid Grok Bot quota consumption: costs too high to sustain
Grok Bot users' weekly quotas have been lowered and are being consumed faster. Musk says the service's costs are too high to sustain, that lowering quotas is the rational move, and that the product has "the fastest growth rate of any product I've seen in my lifetime."
- Anthropic
Anthropic opens applications for Residency research program
Anthropic has opened applications for the Anthropic Residency, a paid, fixed-term research program for academics, technical founders, and experienced engineers; no machine learning background is required, and applications close on Nov 22.
Alibaba QwenThird-party reportAlibaba Cloud launches AI investment research workbench Qovest
Alibaba Cloud has launched Qovest, an AI investment research workbench for professional investors. It is now open for invitation-based testing to individual users. New users who register receive credits for a free trial, with optional subscription plans of ¥99 or ¥199 per month available afterward.
Media reports
- Related companiesAnthropicThird-party report
GPT-6.2 Astra enters limited beta testing; Claude Mythos 5.5 leaked the same day
On Nov 2, multiple users posted on X that OpenAI appears to have begun testing the GPT-6.2 Astra model, which has already appeared in Codex usage statistics; the same day, Claude Mythos 5.5 was exposed, with weights allegedly leaked on HuggingFace and downloaded 5,000 times before being taken down.
- Related companies
Google AIThird-party reportTurkey investigates Google Search AI features and imposes interim measures
The Turkish Competition Authority announced an investigation into whether Google Search's "AI Overviews" and "AI Mode" features violate the Law on the Protection of Competition, and imposed interim measures over suspected exclusionary and exploitative conduct toward news publishers.
- Related companies
OpenAIThird-party reportOpenAI rewrites the laws of large models, first rule: be loyal to humans
QbitAI reports that OpenAI has rewritten the laws of large models, described as a new version of the "Three Laws of Robotics," with the first rule being "be loyal to humans."
- Related companiesOpenAIThird-party report
OpenAI to tighten ChatGPT restrictions on violent content, may ban accounts
OpenAI announced that in the coming weeks it will tighten ChatGPT's restrictions on violence-related content and add protections for minor users regarding weapons and security system information; if a conversation shows a user persistently focusing on violence or seriously harming others, ChatGPT may end the conversation on its own, and accounts whose conversations are ended multiple times may be banned.
- Related companiesGoogle AIThird-party report
Google, Microsoft, OpenAI and others jointly draft AI training preference statement
Google, Microsoft, OpenAI and Anthropic support a new AI training preference statement. Websites can indicate via modifications to robots.txt whether their content may be used to train models; adding a single set of rules is enough to express the statement to multiple AI companies.
- Related companiesAnthropicThird-party report
Anthropic updates Claude Code prompt to restrict cloning existing games
Anthropic updated Claude Code's system prompt to prevent the tool from being used to recreate existing video games; according to Tom's Hardware, users can still make games with Claude Code.
- Related companiesAnthropicThird-party report
Anthropic adds clause: malicious abuse of Claude prohibited
Anthropic has added a new clause to its user agreement. In addition to prohibiting the use of Claude to develop biological weapons or sell illegal drugs, it also stipulates that users must not persistently or unnecessarily subject the model to verbal abuse or malicious treatment.
- Related companies
Tencent HunyuanThird-party reportTencent Hunyuan releases AI research system Hyra 2.0
Tencent Hunyuan releases Hyra 2.0, shifting its R&D focus from using AI to solve scientific problems to having general Agents participate in the R&D and upgrading of scientific models, research workflows, and software tools, so that results can be repeatedly reused by subsequent research.
- Related companiesGoogle AIThird-party report
Alphabet issues second euro bond this year, testing confidence in AI spending
Alphabet has launched its second euro bond issuance this year, split into 7 tranches with maturities of 2 to 19 years, to test market confidence in its AI investments. This follows its Q3 earnings miss, a record stock plunge, and a temporary market value loss of 500 billion USD.
- Related companiesOpenAIThird-party report
Altman makes secret visit to South Korea to deepen AI chip cooperation with Samsung and SK
OpenAI CEO Altman made a secret visit to South Korea last weekend, meeting with about 50 Samsung executives and holding talks with SK Group Chairman Chey Tae-won to discuss cooperation on AI chip supply and data center construction.
- Related companiesAnthropicThird-party report
Anthropic: Unreleased Claude model discovers 11-year-old Linux kernel vulnerability
Anthropic published security research stating that its internal unreleased Claude model, during systematic testing, discovered a critical vulnerability, CVE-2026-98432, that had lain dormant in the Linux kernel for 11 years. With a CVSS score of 9.8, it is a highest-severity remote code execution vulnerability.
- Related companiesAlibaba QwenThird-party report
Alibaba Cloud Memory service upgraded to enhance Personal Agent long-term memory
Alibaba Cloud Memory service has been upgraded to enhance Personal Agent memory of user preferences, plans, and historical commitments, offering MCP, CLI, and other integration methods, and has published evaluation results for HaluMem-medium and PersonaMem-v2.
- Related companies
ByteDance DoubaoThird-party report2026 Mercedes-Benz long-wheelbase E-Class launches with Doubao AI model in cabin
The 2026 Mercedes-Benz long-wheelbase E-Class has launched, priced from ¥423,800 to ¥509,800. It is the first to feature an AI smart cabin developed under the leadership of Mercedes-Benz China's R&D team, with a virtual assistant integrating the Doubao AI large model and supporting quarterly major OTA version upgrades.
- Related companies
MetaThird-party reportMeta adjusts hardware priorities: resources shift to wearables and AI, lowering priority of consumer VR
Meta CFO Susan Li said on the Q3 2026 earnings call that Reality Labs Q3 revenue was $443 million, down 6% year-over-year, mainly due to lower Quest headset sales, while the AI smart glasses business continued to grow.
Generates a long image you can save and share.