At Sapphire in Orlando this May, SAP named Claude as the primary reasoning and agentic capability inside its newly unveiled Business AI Platform.<\/p>\n
Claude now sits behind Joule and the wider Joule agent layer, connected through the Model Context Protocol, drawing on SAP\u2019s Knowledge Graph and what SAP calls Company Memory to ground its reasoning in a customer\u2019s business data rather than a generic prompt.<\/p>\n
Christian Klein framed the move around accuracy: \u201cFor the mission-critical processes of our customers, \u2018almost right\u2019 just isn\u2019t good enough.\u201d Daniela Amodei framed it around trust: Claude was built, in her words, to do the work that keeps a business running (closing the books, rerouting a delayed order, approving an expense) inside systems customers have already invested in and governed.<\/p>\n
Joule Studio, the build environment for custom agents, is explicitly model-agnostic. SAP named Mistral AI and Cohere alongside Anthropic at the same event<\/a>, and described them specifically as sovereign model options for SAP\u2019s cloud infrastructure.<\/p>\n That shows SAP is considering its initiatives in light of the need for digital sovereignty. Claude is the named primary model, but it is not the only model. This matters because \u201cprimary reasoning model\u201d can sound like SAP has handed its entire AI strategy to a single American company.<\/p>\n SAP is choosing Claude as its best current option for the hardest reasoning tasks, while keeping a parallel set of model relationships specifically so that choice doesn\u2019t become a single point of dependency. Whether that parallel structure holds up against real customer and government pressure, particularly from regulated industries and from Europe, is the question worth considering in depth.<\/p>\n The Architecture<\/p>\n The technical case for taking this seriously starts with what Claude gets access to.<\/p>\n Before a Claude-powered agent acts on anything inside a customer\u2019s SAP landscape, it grounds itself against SAP\u2019s Knowledge Graph<\/a>: a structured map of business entities, processes, and relationships specific to that customer\u2019s configuration. That grounding step is the thing separating this from a chatbot bolted onto an ERP system.<\/p>\n A treasury manager can ask Joule to prepare a CFO briefing ahead of a bank meeting, and the agent pulls live data, runs the analysis, and flags financial risk inside the actual SAP environment, connected to the business it\u2019s describing.<\/p>\n NVIDIA\u2019s OpenShell<\/a> provides the runtime, handling technical execution safety, including isolation, resource limits, and policy enforcement.<\/p>\n Joule Studio\u2019s runtime handles a separate layer of business governance: who can deploy which agents, what data they can touch, how an agent\u2019s actions map back to a human role and an approval chain.<\/p>\n SAP has been careful to describe these as two distinct layers rather than one combined safety net, and that distinction will matter to anyone scoping a deployment later.<\/p>\n Access for customers runs through S\/4HANA Cloud with a clean-core approach. Customers still running ECC, or heavily customized on-premise landscapes, don\u2019t get the autonomous functions at all unless they\u2019ve committed to moving the bulk of their landscape to SAP\u2019s cloud ERP.<\/p>\n The 2027 maintenance deadline (2030 for customers with Extended Maintenance) was already pushing migration decisions before this announcement. Now it carries an additional argument: staying put doesn\u2019t just mean an aging platform, but also exclusion from the reasoning layer SAP is building its more innovative products around.<\/p>\n The Sovereignty Fault Line<\/p>\n Anthropic, as things currently stand, offers no EU data residency through its direct API or through Claude Enterprise. Both default to processing in the United States, with Standard Contractual Clauses as the transfer mechanism under GDPR.<\/p>\n A genuine EU-resident path exists, but only through AWS Bedrock\u2019s European regions or Google Vertex AI\u2019s EU endpoints, configured deliberately for that purpose. SAP has not yet published, as far as the public record shows, which of these paths governs Claude\u2019s presence inside the Business AI Platform itself, or whether Claude-in-Joule inherits the same residency gap that applies to Anthropic\u2019s products generally.<\/p>\n On the third of June, the European Commission presented its Technological Sovereignty Package: a Cloud and AI Development Act, a Chips Act 2.0, an open-source strategy, and a roadmap for assessing cloud and AI sovereignty across the region.<\/p>\n SAP deepened its dependence on a US foundation model in the same fortnight the EU was building the legislative apparatus to scrutinize exactly that kind of dependence. The legislation still needs negotiation through Parliament and Council, so nothing here is settled law yet. But the direction of travel is clear, and SAP\u2019s customers in regulated sectors will be the first to be affected by it<\/p>\n German works councils have already started treating this as their territory.<\/p>\n Under \u00a787 Abs. 1 Nr. 6 of the Betriebsverfassungsgesetz, works councils hold a binding co-determination right over technical systems capable of monitoring employee conduct or performance, and German legal commentary is explicit that this applies to AI systems regardless of whether monitoring is the intended purpose: the courts have held that objective capability to monitor is enough to trigger the right.<\/p>\n A Joule deployment that logs usage and activity data will, in most readings, qualify. German trade commentary on the announcement has already framed AI model choice as a negotiating point belonging inside the SAP contract itself, rather than a technical detail settled afterward by IT.<\/p>\n SAP\u2019s actual answer to all of this is the model-agnostic architecture already described: Mistral, Cohere, and presumably others as genuine alternatives for customers who need a different sovereignty profile.<\/p>\n Claude itself has not become sovereign-compliant, and SAP is not claiming it has. SAP has built itself room to adjust its position around a question it can\u2019t yet fully answer.<\/p>\n The Fable and Mythos Suspension<\/p>\n On the ninth of June, Anthropic released Claude Fable 5 and Claude Mythos 5, both built on what the company calls its Mythos-class architecture, sitting above the Opus tier in raw capability.<\/p>\n Fable arrived with safety classifiers built in, designed for general commercial release. Mythos, the same underlying model with those classifiers lifted, stayed restricted to a small group of vetted partners under Project Glasswing: meaning cybersecurity experts and infrastructure providers using its vulnerability-finding capability to patch software flaws before attackers find them.<\/p>\n Three days later, the US government issued an export control directive ordering Anthropic to block access to both models for any foreign national, anywhere, including Anthropic\u2019s own non-citizen staff.<\/p>\n Unable to separate foreign nationals from everyone else in real time, Anthropic disabled both models worldwide rather than risk non-compliance. The government\u2019s stated concern involved a jailbreak technique bypassing Fable\u2019s safeguards. Anthropic\u2019s own account, published the same evening, called it a narrow, non-universal method already replicable on other publicly available models, including OpenAI\u2019s. The company complied while disputing the extent of the finding, and as of this writing, more than a week later, access to both models remains suspended with no confirmed restoration date.<\/p>\n Two things need stating clearly here, because the easy mistake is to collapse them into one.<\/p>\n First: nothing suggests SAP\u2019s production deployment of Claude inside Joule has been affected by any of this. The Claude models actually embedded in the Business AI Platform sit at the Opus tier, not the Mythos tier, and Anthropic has been explicit that the suspension touches Fable and Mythos alone.<\/p>\n Second: A US administration demonstrated, with seven days\u2019 notice from release to suspension, that it can remove an entire model tier from global access overnight, for reasons it didn\u2019t fully explain even to the company building the model. That is the fact worth considering in any conversation about sovereignty, regardless of which specific model sits inside which specific product.<\/p>\n It also complicates the EU\u2019s case in an interesting way, because the complication cuts in the EU\u2019s favor rather than against it.<\/p>\n European regulators arguing for sovereign AI infrastructure have spent two years making an argument about data residency and legal jurisdiction. The Fable suspension hands them a second, sharper argument: that access itself, not just data location, sits at the discretion of a foreign government.<\/p>\n A customer relying on Claude for mission-critical reasoning doesn\u2019t just need to ask where the data sits. They now need to ask whether the model will still be available in the following month, and that is a harder question to answer with confidence.<\/p>\n What This Means for SAP Consulting<\/p>\n The governance and audit-trail mapping that used to be a nice-to-have conversation with a client is now a deliverable in its own right: which scenarios route to Claude, which route to another model, what data each layer can touch, and who can inspect or override an agent\u2019s action before it executes against production.<\/p>\n SAP\u2019s own trade press coverage has started advising customers to ask their account teams and systems integrators to map this explicitly. That advice is worth taking literally and building into a statement of work.<\/p>\n It may also have implications for pricing. Forrester\u2019s research already has roughly a fifth of enterprise SaaS decision-makers naming vendor lock-in as a live commercial concern, and deeper model coupling inside a platform as central as Joule only increases that exposure. A consultancy that scopes a Claude-dependent agentic workflow without naming the lock-in risk to the client in writing is leaving a problem for someone else to discover later. Better to price the risk now than deal with the awkward conversation in eighteen months.<\/p>\n On skills, SAP Learning already has a Joule for Consultants track covering the basics of working alongside the assistant day to day. The work that actually differentiates a consultant through 2027 is more complex than that: involving understanding agent governance well enough to advise a client, knowing what MCP connectivity does and doesn\u2019t expose, and being able to explain to a CFO or a works council, in plain terms, where a given Joule agent\u2019s reasoning occurs.<\/p>\n For clients in Germany and across the DACH region specifically, the works council conversation belongs at the start of a project, not somewhere in the middle once a Betriebsvereinbarung becomes urgent.<\/p>\n There\u2019s also a useful gift buried in SAP\u2019s own admission, made openly by Klein at Sapphire, that earlier Joule releases fell short on accuracy and lost some customer trust as a result. That admission gives any consultant the cover to scope a first deployment as a contained pilot. Clients tend to respect that kind of caution.<\/p>\n Where This Settles For Now<\/p>\n SAP\u2019s integration of Claude into Joule is technically serious: the Knowledge Graph grounding, the governance layering between OpenShell and Joule Studio, and the clean-core prerequisite all suggest a company that has thought through the mechanics.<\/p>\n At the same time, the sovereignty question SAP\u2019s own architecture is built to hedge against remains genuinely open, and the Fable suspension has handed European regulators practical evidence supporting the argument they were already making.<\/p>\n A consultant advising a client through this needs to hold both facts without reaching for whichever one matches the story they heard from SAP\u2019s or Anthropic\u2019s marketing.<\/p>\n The honest answer, right now, is that nobody outside both companies\u2019 legal and compliance teams knows exactly how Claude-in-Joule will be treated if the EU\u2019s sovereignty framework hardens into binding law over the following years.<\/p>\n Three things worth watching are: how the EU\u2019s Technological Sovereignty Package fares once it reaches Parliament and Council, whether Fable and Mythos access returns and on what terms, and what the first independent customer references on Claude-powered Joule deployments actually say once they start appearing later this year.<\/p>\n Until then, the responsible position is to remain informed of developments and not to deliver a settled verdict either way.<\/p>\n If you are an SAP professional looking for a new role in the SAP ecosystem our team of dedicated recruitment consultants can match you with your ideal employer and negotiate a competitive compensation package for your extremely valuable skills, so join our exclusive community at IgniteSAP<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"At Sapphire in Orlando this May, SAP named Claude as the primary reasoning and agentic capability inside its…\n","protected":false},"author":2,"featured_media":41956,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21036],"tags":[36409,36410,36411,36412,21870,9695,21872,21543,21093,36413,21874,21875],"class_list":["post-41955","post","type-post","status-publish","format-standard","has-post-thumbnail","category-sap","tag-ai-advisory-services","tag-anthropic-fable","tag-eu-ai-laws","tag-eu-ai-legislation","tag-ignitesap","tag-sap","tag-sap-ai","tag-sap-anthropic","tag-sap-btp","tag-sap-claude","tag-sap-s-4hana","tag-sap-s-4hana-cloud"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/posts\/41955","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/comments?post=41955"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/posts\/41955\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/media\/41956"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/media?parent=41955"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/categories?post=41955"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/germany\/wp-json\/wp\/v2\/tags?post=41955"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}