Is the insurance "Lobster" just starting its deployment? A major insurance company has already made it available to over 2,000 employees, while most insurers are observing and waiting.

Interface News Reporter | Feng Lijun

The “lobster farming” trend has reached the insurance industry.

Interface News has learned from multiple insurance companies that most have not yet started “lobster farming,” but a few are testing the waters.

Currently, many regions across the country support “lobster farming.” At the same time, the risks associated with “lobster farming” are gradually emerging. On the evening of March 10, the National Internet Emergency Center issued a risk alert regarding the security of the OpenClaw application.

The financial industry is inherently more sensitive to security risks. Media reports indicate that some financial institutions recently received risk alerts and have imposed strict controls on deploying external platforms similar to OpenClaw for security reasons. Should the insurance industry “farm lobsters”? How can they do so safely?

The Future of Insurance “Lobster Farming”

Artificial intelligence technology, represented by generative large models, is flourishing across various industries in the form of AI agents.

The China Artificial Intelligence Industry Development Alliance recently released the “Insurance Industry Intelligent Agent Deployment Roadmap (2025)” which states that, as an important branch of the financial system, the insurance industry features data-intensive operations, diverse products, and multi-faceted services. Intelligent agents have broad development prospects in insurance.

According to McKinsey data, the productivity gains from generative AI in the insurance sector could reach $50 billion to $70 billion, helping improve efficiency and reduce costs in sales, underwriting, claims, customer service, and other areas.

So, what can the currently popular intelligent agent OpenClaw bring to the insurance industry?

“Unlike other general-purpose traditional large models and agents, OpenClaw is an authorized ‘AI digital employee’.” a senior technology executive from a leading insurance company told Interface News. “It can autonomously plan and execute complex tasks, such as data collection, report generation, document processing, and cross-system workflow scheduling. It supports internal network deployment, keeps business data within private domains, and is widely applicable in high-compliance scenarios like finance and healthcare.”

“OpenClaw indeed has broad application potential in insurance, with core value in cost reduction, efficiency enhancement, and service optimization.” said the head of AI projects at Xinghuo Zhiyun Technology (Xinghuo Insurance), an insurance tech and financial service provider. “In internal operations, it can automatically generate meeting summaries, precisely search knowledge bases, reduce clerical work, and allow teams to focus on core business; in customer service, it can extend to intelligent customer support, policy interpretation, and initial claim review, improving response speed and lowering service costs; in risk control and decision-making, it can analyze business data intelligently to assist in underwriting and claims risk identification, enhancing accuracy and processing efficiency.”

“Currently, OpenClaw is in a phase of ‘encouraged pilot exploration’ and ‘safety regulation development.’ It has preliminarily demonstrated high efficiency in real business scenarios and is expected to become a key infrastructure for deepening digital operations within the group,” the senior technology executive added.

Is the “Lobster” Ready for Duty?

From the pilot efforts of these leading insurers, their “lobsters” are already capable of “getting to work.”

“In AI office applications, OpenClaw has integrated with our internal email, meetings, and other key office platforms and capabilities. Most office scenarios such as email handling, scheduling, and work summaries can be completed by OpenClaw. Employees are enthusiastic; currently, over 2,000 staff members are using it, with nearly 200 Skills created.” the senior tech executive said. “In the future, we will gradually achieve full integration of OpenClaw with the company’s platforms, giving employees a true ‘digital clone.’”

Skills refer to the capabilities assigned to the agent, essentially the “user manual” for the “lobster.” They specify the scenarios, tools, tasks, and precautions, and are key to making the “lobster” truly usable by humans. Skills are also one of the barriers that must be crossed from installation to successful deployment of the “lobster.”

An insider from a major cloud provider told Interface News that their internally developed “lobster” for internal networks has not yet explored application scenarios. “It depends on how deeply Skills are researched. Also, many Skills currently have vulnerabilities, and some may execute malicious code.”

The senior tech executive from the leading insurance company said, “Since January 2026, the company has established a Skills R&D zone, accumulating over 120 Skills cases. Among these, the AI role-based assistant performs exceptionally well. It has evolved from a consumer-oriented helper to a team-level AI partner through our extensive skill and team training.”

Xinghuo Insurance’s “lobster” has also been applied in functions like meeting summaries, document retrieval, and short video generation, significantly improving team efficiency.

“We are currently integrating OpenClaw technology into our self-developed AI product ‘Xingju’ and have completed internal deployment,” said the AI project leader at Xinghuo Zhiyun. “To address high installation barriers, the technical team has built a rapid deployment channel, making it easy for non-technical colleagues to use.”

The senior technology executive from the leading insurer added, “In AI operations, our goal is ‘process reengineering + technological evolution,’ with business and technology advancing hand-in-hand to achieve high-level AI replacement in group digital operations.”

“More than 30 scenarios are actively underway in property and life insurance subsidiaries, including intelligent recommendations, customer service avatars, and claims assistants. These not only enhance operational efficiency but also enable personalized marketing, faster responses, and more attentive, efficient services, increasing customer satisfaction and long-term trust,” he said.

How Can the Insurance Industry Safely “Farm Lobsters”?

OpenClaw represents an important trend toward AI inclusiveness and is a powerful tool for digital transformation in insurance. While embracing AI benefits, strict compliance and security must be maintained.

The National Internet Emergency Center warns that, to enable “autonomous task execution,” OpenClaw is granted high system permissions, including access to local file systems, environment variables, external APIs, and extension installation. However, its default security configuration is extremely fragile; once a breach is found, attackers could easily gain full control of the system.

Broadcom’s chief analyst for the financial sector, Wang Pengbo, states that current open-source intelligent agents generally emphasize end-to-end automation, but their operational logic conflicts fundamentally with the transparency, traceability, and human intervention required by financial regulation.

Interface News has learned that most insurance companies have not yet experimented with “lobster farming,” mainly due to security concerns. An insider from an AI agent vendor told Interface News, “Our financial clients probably won’t ‘farm lobsters’—it’s too unsafe.”

Regarding specific concerns about “lobsters,” Yuanyou Financial’s Senior Vice President and AI Chief Officer told Interface News, “Mainly data security and internal-external network isolation. Without external network access, and with strict internal-external network separation, data cannot be exposed externally. Financial institutions promote ‘compliant business intelligent agents’—safe, controllable, auditable, serving financial operations and integration, empowering financial intelligence.”

The senior tech executive from the leading insurer said that OpenClaw has risks such as improper high-permission configurations, supply chain vulnerabilities, and prompt injection attacks. Therefore, the company has issued related security guidelines across the group, allowing only read-only auxiliary scenarios, prohibiting high-permission rewriting in production systems, and forbidding employees from installing or using it privately within the company to ensure safe and orderly use.

“It’s clear that OpenClaw has significant privacy and security issues,” said the AI project leader at Xinghuo Zhiyun. “For insurance data, Xinghuo Insurance has implemented strict restrictions and approval mechanisms to protect user privacy and company assets. We enforce data access levels, strictly control privacy and core data, limit AI capabilities to authorized data, and ensure user privacy and asset security.”

Wang Pengbo believes that for open-source AI agents to truly enter core financial scenarios, six key thresholds must be crossed: 1) explainable and traceable algorithms to eliminate black boxes; 2) clear responsibility and accountability mechanisms; 3) continuous improvement of model reliability and common sense reasoning to reduce errors; 4) compliance with financial data privacy standards; 5) collaborative models balancing open-source spirit and commercial interests to foster an open ecosystem; 6) mandatory human intervention rights and circuit breakers to prevent irreversible risks.

The National Internet Emergency Center recommends that, when deploying and applying OpenClaw, measures such as strengthening network controls, credential management, plugin source management, and ongoing patch and security updates should be implemented.