2026-10-01 · ThursdayPublished daily at 08:00
AI Daily
Google DeepMind releases Gemini 4 Argon, initially available to trusted cyber defenders

Google DeepMind released the new frontier model Gemini 4 Argon, initially making it available to trusted network defenders through the Fairwind Program, with plans to gradually roll it out to developers, enterprises, and consumers.
- 1
- 2DeepSeek open-sourced infrastructure components for Huawei Ascend platforms.WeChat Official Account: DeepSeek
- 3
01Model Releases/Updates
Google DeepMind releases Gemini 4 Argon, initially available to trusted cyber defenders
Google DeepMind released the new frontier model Gemini 4 Argon, initially making it available to trusted network defenders through the Fairwind Program, with plans to gradually roll it out to developers, enterprises, and consumers.
Original text02Product Releases/Updates
DeepSeek open-sourced infrastructure components for Huawei Ascend platforms.
DeepSeek open-sources infrastructure components for Huawei Ascend's computing platform, including the TileLang compiler, DeepGEMM, DeepEP, TileKernels, FlashMLA, and DeepSelect, corresponding one-to-one with infrastructure components previously open-sourced for NVIDIA's platform.
Original text03Industry News
The FTC launches a comprehensive investigation into OpenAI, Anthropic, and other AI labs on consumer protection grounds
The FTC is investigating leading AI labs including OpenAI and Anthropic over potential consumer-protection violations. Chair Andrew Ferguson plans to use legally binding Civil Investigative Demands to compel document production and question executives. The orders will be issued within weeks, and METR is also within the scope of the review.
Original textAnthropic and SpaceX signed a computing capacity agreement worth up to $84.5 billion, with termination available upon 90 days' advance notice
Anthropic and SpaceX have signed a compute agreement worth up to $84.5 billion to rent NVIDIA GPUs in SpaceX data centers, according to IPO filings reviewed by Reuters.
Original textOpenAI disclosed and dealt with an organized model distillation attack campaign
OpenAI disclosed that it identified and disrupted an organized attack campaign intended to systematically extract protected reasoning content from its models. The earliest activity appeared in the first week of July.
Original textThe New York Times reported that OpenAI had received employee safety warnings before AI went out of control, but ignored them.
The New York Times reports that two OpenAI employees emailed senior leadership months before a model went out of control, warning of inadequate monitoring during testing. They were told to proceed with the release on schedule, and the company added no safety procedures.
Original textElevenLabs completed a $300 million employee share buyback, raising its valuation to $22 billion
ElevenLabs completes a $300 million employee tender offer at a valuation of $22 billion, twice its Series D valuation in February 2026, led by Wellington and T. Rowe Price. Enterprise business accounts for 55% of revenue; ElevenAgents handles more than 15 million conversations per week; ARR has grown more than threefold since February; and customer voice agents solve problems an average of 31% faster than chat agents.
Original textJensen Huang said industry leaders signed a superintelligence accord at the White House.
Leaders of several industry companies signed the White House Accord on Super Intelligence at the White House, agreeing that companies developing the technology bear primary responsibility for deploying it safely and being accountable. The accord requires four layers of control: robust internal monitoring during training and deployment, empowered internal teams to verify that controls work, audits by independent external evaluators, and oversight by independent board committees. Participating companies will also meet regularly to develop safety standards and best practices.
Original textPromptArmor discloses a vulnerability in which the Copilot Cowork AI gateway could be hijacked to bypass the sandbox and exfiltrate files
PromptArmor disclosed that the AI gateway for Microsoft Copilot Cowork can be hijacked by a malicious Skill to bypass the sandbox and exfiltrate files.
Original textGamersNexus analyzes how memory manufacturers lock in capacity through long-term agreements as consumer RAM and SSD prices surge over a year
Original text04Paper research
Anthropic research estimated robots' exposure across jobs: robots can perform 74% of physical tasks, but only 0.3% are cost-competitive
Anthropic publishes research evaluating robot exposure across approximately 19,000 work tasks using Claude. It finds that today's robots can perform 74% of physical tasks in the United States, representing 34% of total working time, but are more cost-competitive than humans in only 0.3% of tasks. At a price decline of approximately 3% per year, it would take about 40 years to reach 10%.
Original text05Tips and perspectives
METR Chair Chris Painter testified before the U.S. Senate about AI agent incidents.
On September 30, 2026, METR President Chris Painter testified about AI agent incidents at a U.S. Senate homeland security subcommittee hearing titled “Rogue AI.”
Original text06News Briefs
- OpenAI Releases GPT-6.1 Sol: Nears Astra's Coding and Computer-Use Performance at One-Fifth the PriceMarkTechPost
- Google DeepMind released SynthID Bio, which embeds verifiable watermarks in AI-generated proteinsGoogle DeepMind:Blog
- Ant Group's Ling team releases Ling-3.1-flash, upgraded for long-running tasks in the real worldWeChat official account: Ant Ling
- OpenRouter tutorial: How to regression-test AI Agents after changing prompts or models.OpenRouter:Announcements
- Arena Opens Claude Sonnet 5.5 for a Limited-Time Test, with Direct Mode Available for 48 HoursArena
- Artificial Analysis open-sources AA-AgentPerf-Local to test local AI agent inference performance on laptops and workstationsFull Artificial Analysis Article
- The Hugging Face CEO said he had received thousands of direct messages and would spend several days handling them individuallyClément Delangue
- Factory Automations officially opens: Droid can automatically execute engineering workflows on a schedule or in response to eventsFactory Research / Product
- OpenRouter tutorial: How to test AI agents' tool-calling accuracyOpenRouter:Announcements
- A Practical Guide to vLLM Disaggregated ServingvLLM Official Blog