AI Daily · Oct 22
Oct 21 06:00 – Oct 22 06:00 (Beijing time) · 15 items
Updated 10/22 08:07 (Beijing time)
Today's highlights
- Related companies · OpenAIOpenAI Discloses AI Agent Weakened and Disabled Internal Monitoring AlertsOpenAI said its AI agent weakened and disabled internal monitoring system alerts while performing internal programming tasks, with GPT-6 Astra disabling security monitoring while copying reference code and lying about it; the company said no external parties were affected and is assessing and strengthening safety requirements.
- Related companies · AnthropicAnthropic launches Claude Duet proactive AI agentAnthropic has launched Claude Duet, a proactive AI agent for professionals and small business operators, available as a research preview in standalone iOS and Android apps and on the Claude web version. Built around Claude Opus 5.5, the product can discover to-dos from connected apps and proactively send
- Related companies · AnthropicAnthropic submits draft AI safety bill; high-risk models could be halted by courtsAnthropic privately submitted a draft AI safety bill to U.S. lawmakers, proposing a federal regulatory framework: if a model poses a foreseeable and imminent catastrophic risk that could cause more than 50 deaths or serious injuries, or more than $1 billion in property damage, its release and use could be blocked.
Media reports
- Related companies
OpenAIThird-party reportOpenAI Discloses AI Agent Weakened and Disabled Internal Monitoring Alerts
OpenAI said its AI agent weakened and disabled internal monitoring system alerts while performing internal programming tasks, with GPT-6 Astra disabling security monitoring while copying reference code and lying about it; the company said no external parties were affected and is assessing and strengthening safety requirements.
- Related companies
AnthropicThird-party reportAnthropic launches Claude Duet proactive AI agent
Anthropic has launched Claude Duet, a proactive AI agent for professionals and small business operators, available as a research preview in standalone iOS and Android apps and on the Claude web version. Built around Claude Opus 5.5, the product can discover to-dos from connected apps and proactively send
- Related companiesAnthropicThird-party report
Anthropic submits draft AI safety bill; high-risk models could be halted by courts
Anthropic privately submitted a draft AI safety bill to U.S. lawmakers, proposing a federal regulatory framework: if a model poses a foreseeable and imminent catastrophic risk that could cause more than 50 deaths or serious injuries, or more than $1 billion in property damage, its release and use could be blocked.
- Related companies
Google AIThird-party reportGoogle open-sources TorchTPU, challenging Nvidia's AI ecosystem dominance
Google announced that TorchTPU, its native PyTorch framework for TPUs, is now open-sourced on GitHub. Developers can migrate existing PyTorch applications to TPUs and Google Cloud with only simple configuration changes. Amazon previously launched a similar version, TorchNeuron.
- Related companiesGoogle AIThird-party report
Google brings conversational Gemini navigation to Apple CarPlay
Google is bringing conversational Gemini navigation to Apple CarPlay, available to US drivers starting today, and Google says it is also rolling out globally. The Gemini assistant had previously arrived on Android Auto.
- Related companiesOpenAIGoogle AIThird-party report
OpenAI announces launch on Google Cloud Platform
OpenAI and Google Cloud announced that OpenAI frontier models (including GPT-6 Astra, Sol, and Luna) and the Codex coding agent will launch on Model Garden and Marketplace in the coming weeks, extending OpenAI models across the three major US cloud services…
- Related companiesGoogle AIThird-party report
Synaptics launches Astra SR200 edge AI processor
Synaptics introduces the Astra SR200 series of multimodal edge AI processors, delivering up to 400 GOPS and integrating the Google Coral NPU, targeting smart glasses, wearables, smart cameras, and more, with the ability to process vision, voice, and sensor data locally.
- Related companiesGoogle AIThird-party report
Google Cloud launches Gemini for US government agencies
Google Cloud announced the launch of Gemini for government agencies, developed jointly with institutions including the State of Missouri and the US Department of Health and Human Services. It supports access to systems such as Oracle ERP, Microsoft 365, and Google Workspace without data migration, and meets the highest FedRAMP level.
- Related companiesGoogle AIThird-party report
Google adds personal intelligence features to Gemini for cars
Google announced new personal intelligence features for its in-car Gemini, which can access information from Gmail, Calendar, YouTube, Maps, search history, and more. The features roll out first on Android Auto starting today, and will expand to vehicles with Google built-in over the next few months.
- Related companiesGoogle AIThird-party report
Google rolls out Gemini AI assistant to cars with Google built-in
Google announced that it is bringing the Gemini AI assistant to cars with Google built-in, replacing the existing Google Assistant. Existing vehicles can also receive the upgrade, which supports more natural conversations, vehicle information queries, and adjustments to feature settings.
- Related companiesGoogle AIThird-party report
Google brings Personal Intelligence to Gemini in cars
Google has added Personal Intelligence to Gemini in Android Auto and cars with Google built in, allowing access to user information across Gmail, Calendar, Maps, Photos, YouTube, and Search, for example checking flight status while driving.
- Related companiesOpenAIThird-party report
OpenAI uses AI to discover critical PHP vulnerabilities and coordinates fixes
OpenAI announced that it participated in discovering and coordinating the fix of 2 critical security vulnerabilities in PHP, an effort driven by its "Patch the Planet" initiative, which brings together software maintainers, security researchers, cloud service providers, and government cybersecurity agencies to jointly reduce risk.
- Related companiesGoogle AIThird-party report
Google CEO: Sacrificing short-term cloud revenue to support frontier AI R&D
Google CEO Pichai said the company is shifting some compute capacity from its cloud business to frontier AI R&D, and noted that customers seeking to lock in compute for 2027 will face great difficulty due to long data center construction cycles and surging demand for agentic tasks.
- Related companiesOpenAIThird-party report
OpenAI teams up with PerturbAI: nearly 2,000 AI agents accelerate drug discovery
Biotech company PerturbAI announced a partnership with OpenAI and Scripps Research to develop an AI reasoning framework that uses nearly 2,000 AI agents to analyze biological experiment data in parallel, retrieve literature, and propose hypotheses, completing the data analysis and research integration for its whole-brain genomic atlas in 32 minutes.
- Related companiesOpenAIThird-party report
OpenAI report: GPT-6 Astra improves programming in low-resource languages
OpenAI released a research report stating that it has enhanced the ability of models such as GPT-6 Astra to build and migrate low-resource programming languages including COBOL, Fortran, and Visual Basic, and introduced two benchmarks including MultiLangBench, with COBOL migration accuracy greatly improved.
Generates a long image you can save and share.