SAP Vulnerabilities and the Risk of Rushing AI Features

Serious SAP security vulnerabilities disclosed this year raise an uncomfortable question about priorities. Vendors are racing to add AI assistants to enterprise software while the platforms underneath still contain flaws that expose finance, supply chain and personnel data. Before adding clever new features, customers should be sure the foundation is secure.

What the flaws allowed

Researchers showed that attackers could reach sensitive business records and, in some cases, run commands with high privileges on the server. These systems sit at the centre of a company, so a single weakness can expose invoices, payroll, supplier bank details and production plans. Because many installations are customised and heavily integrated, patches are often delayed for months, leaving a long window for attackers.

Why AI features raise the stakes

An assistant that summarises reports or answers questions needs broad read access to the same data. If the platform has a gap, the assistant becomes another route to that data. New components also bring new risks such as prompt injection, where hidden text in a document tricks the assistant into revealing information or performing an action. Each integration point must be treated as a potential entry.

Patch the core platform first

Apply security notes on a regular schedule, starting with those that rate as critical or are known to be exploited. Maintain a test system that mirrors production so patches can be validated quickly. Remove unused components and accounts, and restrict administrative access to named people with multi-factor sign-in. A patched, minimal platform is the best base for any new feature.

Limit what AI tools can see and do

Give assistants the minimum access required for a specific job, ideally read-only and scoped to one department. Log every query and response, and review the logs for unusual requests. Keep sensitive fields such as bank details out of the data the assistant can reach. Test for prompt injection before launch using realistic malicious documents.

Governance that keeps pace

Create a short approval process for any new AI feature in core systems, involving security, legal and the business owner. Document what data is used, where it is processed and who is accountable. Reassess after each major vendor update, because capabilities and permissions often change silently.

Key takeaways

• What the flaws allowed: Researchers showed that attackers could reach sensitive business records and, in some cases, run commands with high privileges on the server. These systems sit at the centre of a company, so a single weakness can expose invoices, payroll, supplier bank details and production plans.
• Why AI features raise the stakes: An assistant that summarises reports or answers questions needs broad read access to the same data. If the platform has a gap, the assistant becomes another route to that data.
• Patch the core platform first: Apply security notes on a regular schedule, starting with those that rate as critical or are known to be exploited. Maintain a test system that mirrors production so patches can be validated quickly.
• Limit what AI tools can see and do: Give assistants the minimum access required for a specific job, ideally read-only and scoped to one department. Log every query and response, and review the logs for unusual requests.
• Governance that keeps pace: Create a short approval process for any new AI feature in core systems, involving security, legal and the business owner. Document what data is used, where it is processed and who is accountable.

Frequently asked questions

Should we avoid AI features in enterprise software?

Not necessarily. Use them where the benefit is clear and the platform is patched, monitored and tightly scoped.

How fast should critical SAP patches be applied?

Within days for exploited flaws, otherwise within the next monthly cycle.

Final thoughts

Speed is not a security strategy. Secure and patch the core first, restrict what AI features can reach, and keep a clear approval process. That order protects the data that matters most.