SAP Moves to Block OpenClaw and Other Unauthorized AI Agents
The Takeaway
- SAP moves to block customers from using unauthorized AI agents.
- External AI agents threaten SAP licenses and Joule AI product sales.
- Longtime customers Mercedes-Benz reduced its SAP instances 40% in recent months.
Some enterprise software companies are worried enough about customers using AI agents to access their data that they’re planning to install tollgates to their apps. And then there are software companies like SAP.
Last month, the $200 billion German firm published a policy document for customers, suggesting it would ban them from using external AI agents to access data they store in SAP apps without the company’s official endorsement. SAP didn’t specify which AI agents it was targeting with the policy. Its warning could apply to AI agents developed by rivals such as Salesforce and ServiceNow, as well as tools like OpenClaw.
The move amounted to an escalation in the brewing data wars between incumbent business app makers and the AI customers are using to get work done inside those apps or to move data out of the apps so the AI can analyze it.
SAP said customers could face “throttling, suspension, or termination of access” if they don’t comply with the new rule. (The vast majority of Fortune 500 companies use SAP for managing supply chains, manufacturing or human resources.)
In a call with analysts in April, SAP CEO Christian Klein said that customers shouldn’t worry about the new policy and that the company wants “to have an open platform.” SAP, he said, has to take new measures to slow down “mass data requests” that could cause “performance issues” with applications if they aren’t kept on a leash.
He also said SAP will “protect” the “intellectual property” it weaves into its products to help customers understand their business processes and the relationships between different kinds of data, such as which customer placed a specific order and which warehouse handled it. External AI agents need to access such information, defined in industry parlance as semantic models or ontology, to analyze a customer’s data, according to a former SAP manager.
SAP’s move is a sign that the company is feeling threatened by the AI tools customers are using to do their work. In theory, agents could lessen software customers’ need to buy more licenses based on the number of workers who access business apps. External AI agents also could threaten sales of SAP’s main AI product, an assistant called Joule, which aims to help customers speed up the collection of payments from customers, analyze contracts, reorder products and handle customer service inquiries.
Some SAP customers feel Joule’s capabilities lag those of Claude and Microsoft Copilot agents, which is why they’ve been using those AI tools to pull data out of SAP, said the former SAP manager. SAP customers can use AI agents for tasks such as determining why their customers’ purchasing patterns have changed or figuring out the reason for a delayed product shipment.
Mercedes Cuts Down on SAP
In another example of new pressures SAP faces, longtime customer Mercedes-Benz says it has reduced its SAP instances—copies of the software employees use—40% to 600 in recent months, according to Chief Information Officer Katrin Lehmann. The carmaker used its own AI models and those from frontier labs—a term that usually refers to OpenAI and Anthropic—to clean up its data, including in SAP apps, to help its other AI tools work better.
People who represent SAP customers find its new AI policy concerning.
“SAP is changing the rules midgame on how customers can get their own data out of their own SAP systems, and that hits every [chief information officer] who built an AI road map assuming open access to” data in SAP’s apps, said Mietske van Ravesteijn, a commercial lead at Redress Compliance, which negotiates software licensing agreements.
Firms such as SAP have plenty of reasons to be concerned about the AI agent threat from OpenAI and others. During an investor presentation earlier this year, for instance, OpenAI leaders said they expected the company’s future products targeting businesses to replace software from firms including Salesforce, Workday, Adobe and Atlassian, The Information reported.
SAP’s shares are down around 28% this year and more than 40% over the past 12 months, largely due to investors’ fears about how AI could affect its business. AI threats don’t appear to have impacted its revenue growth and profits so far. Similarly, Atlassian shares soared last week after the software firm reported a big jump in revenue, suggesting AI hasn’t hit it hard either, though its stock price is still down 40% this year.
A spokesperson said in a statement that “SAP is working with its partners to define access” for AI agents and that the company already allows customers to use agents powered by Microsoft, Google, Amazon and IBM to access their data in SAP apps. They said SAP has “great engineering collaboration with different frontier AI companies, including Anthropic, and we will be publishing more agentic integration architectures soon.”
That suggests SAP may be near a deal to allow Anthropic’s Claude Code or Cowork customers to access their data in SAP. (Spokespeople for Anthropic and for OpenAI, which is developing its own product for creating AI agents that access enterprise apps, didn’t respond to requests for comment.)
On the other hand, the spokesperson said, “open-source AI agent harnesses” like OpenClaw have “documented security and enterprise readiness risks that must be assessed before connecting them to productive enterprise systems.”
‘Short-Term Pain’
Klein, on an earnings call last month, said SAP and its software peers are likely to experience “short-term pain” as they figure out how to develop and sell AI products.
Some of SAP’s peers are striking a similar harsh tone toward external AI agents. Executives at Workday and HubSpot have indicated they’re not happy with AI agents tapping data in their systems. Workday, Salesforce and ServiceNow have said they would essentially erect tollgates to monetize such traffic, a departure from the free flow of data most software companies have allowed between their applications and others in the past.
SAP leaders have long sought to protect the company’s status as a central place where customers store their data. Several years ago, these leaders, including Klein, became concerned that some customers were using other software firms’ analytics tools to handle data they stored with SAP, according to the former manager. The same issue has arisen in the generative AI era, as customers now use the latest tools from AI firms to analyze their SAP data, this person said.
AI agents pose a similar kind of threat. Many SAP customers build custom software to map out the multiple steps needed to complete tasks like approving expense reports, processing invoices and bringing on new employees. Customers can’t take this code with them if they switch to another provider, which helps SAP keep them in the fold. But agents typically tap applications from multiple providers to carry out their work, which means customers don’t have to rely as much on SAP applications.