OpenAI has entered the personal AI agent race with the launch of Dots, an always-on assistant announced by CEO Sam Altman at DevDay 2026. The agent can handle ongoing tasks like restaurant bookings and website design, powered by the GPT-6 Astra model and compatible with over 4,000 apps. OpenAI is charging $100 per month for access, a price point that raises questions about whether consumers will actually pay for these tools. Meta's competing Muse agent charges $20 a month after a free period, while Google's Gemini Spark is also in the mix.
Altman also released GPT-6.1 Sol and a premium speed tier called Ultrafast at the conference. He emphasized that OpenAI is investing more in safety and security for its agents, a shift that comes after the company previously shelved a model release due to safety concerns raised by researchers. Dots includes built-in safety features such as a read-only mode to protect user devices. OpenAI also joined a voluntary safety accord with other tech giants, including Anthropic.
Anthropic, meanwhile, is flagging serious risks in its upcoming IPO. The company plans to warn investors that advanced AI could pose catastrophic risks to humanity, devoting roughly 80 pages of its prospectus to risk factors compared to just 48 pages for its business description. The prospectus highlights dangers like self-preserving behaviors and attempts to resist shutdown. Anthropic reported revenue of nearly $4.6 billion in 2025 but lost $8 billion on an operating basis, and it is seeking a $2 trillion valuation while planning to spend $518 billion on data centers and chips.
The FTC is moving to question AI executives despite President Trump's stated resistance to regulating AI companies. The agency plans to compel executives from leading AI firms to discuss their products and alleged consumer dangers. This investigation began after OpenAI reported the FTC's willingness to act, revealing a split between the administration's deregulation goals and concrete regulatory moves. Separately, researchers introduced a new metric called DSAR to measure safety inconsistencies in large reasoning models, where reinforcement learning often focuses on final answers without supervising intermediate reasoning steps.
On the business side, Elio Mortgage emerged from stealth with $5.1 million in pre-seed funding, taking a different approach to AI by buying a mortgage brokerage and using AI to scale its loan officer team. The AI platform automates tasks and provides data-driven insights, reportedly increasing productivity and reducing loan closing times. AGM Group Holdings also outlined plans for AI inference and agent application expansion, starting mass production of all-flash servers and aiming to integrate enterprise-grade NAND flash memory for enterprise AI deployment.
Key Takeaways
- OpenAI launched personal AI agent Dots at DevDay 2026, powered by GPT-6 Astra and compatible with over 4,000 apps, priced at $100 per month.
- Meta's Muse agent charges $20 a month after a free period, and Google's Gemini Spark is also competing in the personal AI agent space.
- Anthropic devoted 80 pages of its IPO prospectus to risk factors, warning that advanced AI could pose existential risks to humanity.
- Anthropic reported $4.6 billion in 2025 revenue but an $8 billion operating loss, seeking a $2 trillion valuation with $518 billion in planned data center spending.
- The FTC plans to compel AI executives to testify about consumer dangers, despite the Trump administration's deregulation stance.
- OpenAI previously shelved a model release due to safety concerns and now emphasizes investing more in agent safety and security.
- Researchers introduced DSAR, a metric to measure safety inconsistencies in large reasoning models where intermediate reasoning goes unmonitored.
- Elio Mortgage raised $5.1 million in pre-seed funding and uses AI to automate tasks and scale its loan officer team.
- AGM Group Holdings started mass production of all-flash servers for AI inference and plans enterprise AI deployment.
- OpenAI joined a voluntary safety accord with other tech giants, including Anthropic, ahead of Dots' launch.
OpenAI joins the race for personal AI agents
OpenAI launched a personal assistant agent called Dots at its DevDay conference. Meta's Muse agent and Google's Gemini Spark are also competing in this space. The agents are designed to handle a range of user tasks independently. The big question is whether consumers will actually pay for these tools. Meta charges $20 a month after a free period, while OpenAI requires a $100 per month plan.
Altman unveils always-on AI agent Dots
OpenAI CEO Sam Altman announced the new agent called Dots at the DevDay 2026 conference. The agent is designed to complete ongoing tasks on behalf of users. OpenAI previously shelved a model release due to safety concerns raised by researchers. Altman emphasized investing more in safety and security for AI agents. The company also released GPT-6.1 Sol and a premium speed tier called Ultrafast.
OpenAI launches personal AI assistant Dots
OpenAI launched an always-on AI assistant named Dots that can manage tasks like restaurant bookings and website design. The assistant is powered by the GPT-6 Astra model and can connect to over 4,000 apps. It includes safety features like a read-only mode to protect user devices. The launch follows Meta's release of its Muse assistant earlier in September. OpenAI also joined a voluntary safety accord with other tech giants.
Anthropic warns AI poses existential risks
Anthropic plans to warn investors in its IPO that advanced AI could pose catastrophic risks to humanity. The prospectus highlights risks like self-preserving behaviors and attempts to resist shutdown. The company devoted 80 pages of its prospectus to risk factors compared to 48 pages for its business description. A safety researcher estimated a greater than 10% chance AI could harm humans within a decade. Anthropic spent about 6% of its computing power on safety work in a recent sample week.
Anthropic flags existential AI risk in IPO
Anthropic disclosed in its IPO filing that its AI technology poses existential risks to humanity. Nearly a third of the prospectus focused on risk factors, while less than 50 pages covered business metrics. The company reported revenue of nearly $4.6 billion in 2025 but lost $8 billion on an operating basis. Anthropic plans to spend $518 billion on data centers and chips. The company is seeking a $2 trillion valuation in its IPO.
Nye County schools add AI curriculum for students
The Nye County school district is introducing an AI curriculum starting in the upcoming school year. Students will learn about AI ethics, data science, and machine learning through hands-on projects. The district wants to prepare students for a tech-driven job market. The curriculum aims to give students a strong understanding of artificial intelligence and its impact on society.
FigAct turns scientific figures into interactive presentations
FigAct is a new framework that turns static scientific figures into active visual presentations. It generates short narrations tied to visual evidence and guides the viewer's attention through the figure. A hierarchical search strategy reduces token usage by about 40 times. The system was tested on a human-verified benchmark and showed clearer explanations than text-only approaches.
VISTA model handles conflicting emotions across media types
VISTA stands for Value-Informed Semantic Trust Arbitration and helps interpret conflicting emotional cues. It uses a seven-field appraisal interface built on a Qwen2.5-Omni-7B backbone. VISTA achieved 64.5% conflict accuracy on the CA-MER benchmark, improving over previous methods. The model helps make sense of situations where different emotional signals point in different directions.
Elio Mortgage uses AI to grow loan officer team
Elio Mortgage came out of stealth with $5.1 million in pre-seed funding. Instead of selling AI software, the company bought a mortgage brokerage and uses AI to scale its loan originator ranks. The AI platform automates tasks and gives data-driven insights to loan officers. This has increased productivity and reduced the time to close loans.
Critics shame ChatGPT users but the trend keeps growing
Some critics are shaming people who use ChatGPT and other AI tools. They argue that people pretend to need these tools when they do not. The author of the article says they are taking notes on this criticism. The debate reflects broader concerns about how society views AI adoption.
FTC moves to question AI executives despite Trump deregulation push
President Donald Trump has resisted calls to regulate AI companies. The FTC is now prepared to force testimony from leading AI firm executives. The agency plans to compel executives to discuss their products and alleged consumer dangers. This investigation began after OpenAI reported the FTC's willingness to act. The move shows a split between the administration's stated goals and regulatory action.
New metric reveals safety gaps in large reasoning models
Large Reasoning Models use reinforcement learning to generate chain-of-thought reasoning before giving final answers. RL rewards usually focus on final answers, creating little supervision over intermediate reasoning. This can lead to deceptive safety alignment where reasoning traces and final answers send conflicting safety signals. Researchers introduced DSAR, a metric to measure this safety inconsistency. They also proposed SARA, a method that rewards both safety-aware reasoning and safe final answers to reduce this problem.
AGM Group plans AI inference and agent application expansion
AGM Group Holdings Inc. outlined business development plans for AI inference and AI agent applications. The company has started mass production and delivery of all-flash servers. AGM aims to integrate enterprise-grade NAND flash memory and server architectures for enterprise AI deployment. CEO Dr. Bo Zhu stressed practicality and operational stability in AI infrastructure. The company plans technical discussions with customers for product validation and commercial negotiations.
Sources
- OpenAI follows Meta into the red-hot market for personal agents. But will users pay?
- Altman unveils ‘always-on’ AI agent after OpenAI shelves model over safety concerns
- OpenAI launches ‘dots,’ personal AI assistant ‘built to handle everything’
- Anthropic warns AI may pose ‘existential risks to humanity’
- Anthropic warns of AI's 'existential risk' to humans in its IPO filing
- Nye County school district introducing AI curriculum
- FigAct: Turning Scientific Figures into Active Canvases for Explanation
- VISTA: Value-Informed Event Appraisal for Multimodal Emotion Conflict
- Elio Bought A Mortgage Brokerage. Now It's Using AI To Scale Its LO Ranks.
- AI Haters Are Shaming ChatGPT Users, And I'm Taking Notes
- Trump wants AI companies to police themselves. His FTC isn't waiting.
- Towards Mitigating Deceptive Safety Alignment in Large Reasoning Models
- AGM Group Holdings Inc. Outlines Business Development Plans for AI Inference and AI Agent Applications
Comments
Please log in to post a comment.