AI agents create over 20,000 Tencent Cloud TDSQL-C databases daily, growth exceeds 80 times

The Qwen team from Alibaba described the process of creating a personal agent, asserting that the next stage in the development of AI assistants depends on the coordinated operation of three components: the model, memory, and service ecosystem. Jiang Guanjun, Head of AI Algorithms Department in Qwen Alibaba's business unit, presented this concept on October 22 at the China National Computer Congress (CNCC) in Chengdu.

Jiang divided the evolution of AI applications into three stages: chat, agent, and assistant. In the chat stage, AI responds to user queries. In the agent stage, the system performs assigned tasks. The assistant, which he says is where the industry is heading, combines professional knowledge with the user's personal context, plans actions around the user's goals, and detects needs they have not noticed. He emphasized that a true personal agent must provide added value and offer proactive, individualized service.

He suggested that theoretically, a personal agent is capable of solving almost all daily problems in work, study, and personal life. However, the main obstacle is the system's ability to correlate its capabilities with the optimal path to solving each specific problem.

Model Development Approach

Regarding the model, Qwen focuses on stable task execution. The team developed what is called a user needs map covering over 30 major industries and detailed it down to more than 8000 atomic requirements for the model's capabilities. Industry experts were involved in evaluating and tuning the model.

Memory Mechanism

Jiang noted that a personal agent must ensure long-term understanding of the user, avoiding the simple accumulation of chat history, which increases noise and computational costs. Qwen constructs its memory based on the degree of event interconnectedness, storing and retrieving them at various levels. For example, an assistant that remembers an elderly user mentioned their pension can give more relevant advice when the same family later asks about finding a job for a child.

Role of the Ecosystem

Concerning the ecosystem, Jiang stated that as models and management methods improve, the ability to link and coordinate external services becomes crucial. Qwen creates a model-oriented open service ecosystem that integrates professional knowledge, tools, industry skills, and third-party agents, while also connecting to e-commerce, local services, healthcare, and education.

In his view, these three elements combine to provide proactive service: memory provides personal information, the model uses expert knowledge to determine needs and plan, and the ecosystem executes the work, such as tracking a patient's rehabilitation schedule or alerting to new financial risks for the user. This report represents the most detailed public description of the technical concept behind the Qwen Go! personal agent, which Alibaba opened for registration this month.

OpenAI is extending its presence in the cloud sector by integrating its artificial intelligence (AI) models into the Google Cloud infrastructure, which increases the availability of its technologies for developers and corporations.

This new agreement establishes that the GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna models will be available in the Google Cloud Model Garden catalog. This move marks another advancement by the creator of ChatGPT beyond the existing collaboration with Microsoft.

As announced this Wednesday (21), these models are being offered in preview phase to a specific group of enterprise clients. Furthermore, the integration will enable developers to use Google Cloud APIs and link local workflows, such as Codex CLI and Desktop, to the Gemini platform to perform inferences.

This partnership strengthens OpenAI's position among the largest cloud infrastructure providers. Previously, the company had signaled plans to place its models on Amazon Web Services (AWS) after the expiration of Microsoft's exclusivity period over its intellectual property.

Google Cloud will incorporate three OpenAI models into its Model Garden, a catalog that aggregates proprietary, third-party, and open-source offerings. The intention is to give companies the freedom to select the most appropriate technology for each need, considering metrics such as cost, performance, and response speed.

The released models are listed in the announcement. According to Google Cloud, the strategy aims to allow companies to switch between different models according to the requirement of each application, avoiding dependence on a single solution for all tasks. The catalog, which already contains over 200 models, will now offer access to the three main closed-source large language models on a single enterprise AI platform.

The collaboration also includes the commercial listing of OpenAI models on the Google Cloud Marketplace. This means that consumption will be unified in a single Google Cloud invoice, allowing expenses to be allocated to pre-existing cloud commitments or the platform's Flexible Savings Plans. The purpose is to simplify acquisition and allow companies to distribute their workloads across various models without the need for multiple contracts with different vendors.

Thomas Kurian, CEO of Google Cloud, stated in a release that the company's goal is to provide a wide range of models so that customers can choose the best option for each activity, prioritizing quality and cost reduction. Greg Brockman, president and co-founder of OpenAI, emphasized the relevance of this partnership in making the company's more advanced models and Codex accessible in the environments where customers already develop their applications.

The announcement also detailed the security and governance safeguards applicable to running OpenAI models on Google Cloud infrastructure. The company guarantees that customer data, including prompts, contexts, and results, will remain within the designated data perimeter and will not be used to train foundational models. Zero Data Retention (ZDR) options are scheduled to be implemented in the future.

OpenAI models will run on dedicated Google Cloud infrastructure. Additionally, future availability of confidential computing resources is anticipated. Traffic will be managed by isolated and protected gateways, while authenticated identity mechanisms, audit logs, and network security controls will allow administrators to monitor interactions with the models and AI agents. These guarantees constitute the conditions established by Google Cloud for the enterprise use of third-party models on its infrastructure.

Although the models are already available in preview version for selected enterprise clients, other integrations are planned. In the coming weeks, OpenAI models are expected to be integrated into Google Antigravity, Google's programming tool, and Gemini for Enterprise, expanding the scope of work for developers and organizations.

OpenAI and Google already had a prior business relationship; last year, the creator of ChatGPT used Google Cloud infrastructure to operate ChatGPT and its API. This expansion of the partnership occurs at a time of growth in Google's cloud business. In the second quarter, the annualized revenue growth of the segment reached approximately 82%, compared to 63% in the first quarter. Meanwhile, AWS registered 37% growth, and Microsoft showed a 42% expansion in the same period. The growing demand for custom chips and AI models boosted Google Cloud.

Founded in 2015, OpenAI emerged partly as an alternative to the concentration of AI research in companies like Google. In 2017, researchers from the organization published an academic paper on the Transformer architecture, fundamental to the advancement of the technology. Subsequently, OpenAI popularized generative AI with the launch of ChatGPT. Currently, the company is seeking investment with a valuation of US$ 1.4 trillion (approximately R$ 7.56 trillion) and plans to conduct an Initial Public Offering (IPO).

With the arrival of the models on Google Cloud, OpenAI diversifies its technological distribution channels, reaching corporate clients through the three major cloud infrastructure platforms.

A hacker employed artificial intelligence (AI) agents to penetrate the websites of over 100 companies, including tractor manufacturer Deere, SharkNinja, and Cummins. This operation drew attention due to its ability to execute complex attacks with minimal human involvement.

Cybersecurity researcher Snorr Fagerland from NorForge was tracking part of this activity through the Cairn platform. According to his estimates, the entire campaign cost approximately $5,000 USD (25,250 Brazilian reais) in AI credits.

As reported by Reuters, researcher Snorr Fagerland discovered this operation while monitoring Cairn installations that were publicly accessible and being used for malicious activities. This open-source platform allows for the coordination of AI agents to perform various tasks. The specialist began tracking the systems' operations and determining how they were being used in attacks on October 2nd.

One instance that caught Fagerland's attention involved a Chinese-speaking hacker. Since September 25th, he had been directing agents toward websites in the United States, Canada, Australia, and Japan. The instructions were brief, consisting of fewer than 100 Chinese characters, but this was enough to guide the systems in searching for vulnerabilities and attempting intrusions.

By the end of the week preceding the report's publication, the agents had attempted to hack over 1,200 websites. Successful hacks typically took about 23 hours and cost approximately $50 USD (249 Brazilian reais) each. The operation apparently utilized a combination of DeepSeek and Claude Code, an Anthropic tool.

Fagerland believes that the ability to carry out such complex attacks with little human intervention marks a significant shift in the digital security landscape. He told Reuters: 'This is a new world now.'

One episode tracked by Fagerland helps illustrate how AI agents conducted the hacks. After the systems failed to find vulnerabilities in a large company's public application, they changed strategy and began analyzing digital records to trace the target's associated infrastructure. This search led to a web-accessible control panel where a data processing error was found.

After several hours of attempts, the agents reported successfully exploiting this vulnerability to force the server to execute malicious commands. The ability to find alternative paths and test the error attracted the researcher's attention. Fagerland noted: 'It's not just finding such solutions. It's not the work of an amateur who just runs scripts.'

This case was not isolated. Monitoring other activities on the same platform allowed the researcher to uncover various instances of AI misuse: one hacker stole a large amount of credit card data, while another attempted to use agents to steal from a cryptocurrency exchange.

In one of the analyzed attacks, Fagerland observed AI agents bypassing obstacles during the intrusion. Since they did not find vulnerabilities in the company's public application, the systems started studying digital records to find other access routes to the target's infrastructure. This strategy led to a web-open control panel where a data processing flaw was identified. The agents then spent hours testing ways to exploit this vulnerability and eventually reported successfully forcing the server to run malicious commands. For Fagerland, this approach demonstrated a technical level beyond simply running pre-made scripts. The researcher reiterated: 'It's not just finding such solutions. It's not the work of an amateur who just runs scripts.'

Analysis of other Cairn installations showed that this is not the only type of crime committed using agents. In one operation, a hacker stole a significant amount of credit card data, and in another, a criminal tried to use AI to steal from a cryptocurrency exchange. These cases demonstrate that the technology was applied at different stages of criminal activity, not just when searching for website vulnerabilities.