AI in the Space Engineering: Secure Information Starts With Clear AI Boundaries

Space projects generate large amounts of engineering information throughout their lifecycle. AI offers new ways to work with that information, from finding relevant records to supporting engineering analysis. These capabilities can save engineers time throughout a space project. AI is also becoming part of the wider space engineering landscape. ESA, for example, is exploring AI across mission operations through its AI and Data Foundation, with attention to data management, governance and assurance.

This article explores AI within space engineering from a practical perspective. It looks at what information AI should or shouldn’t be able to access, what AI can do with that information and how AI can work alongside a controlled engineering environment.

Throughout the article, practical questions helps you thing about AI use cases and decide where AI usage for your project belongs.

Define what AI can see

Giving AI access to engineering information creates an information boundary. The organisation decides which documents, records and systems sit inside that boundary and which stay outside it.

Ask yourself: what information does the AI need to see to perform this task?

Identify which engineering information AI can access

Space programmes contain large amounts of controlled engineering information. Requirements, quality records, test evidence and mission information all support engineering decisions. Their source, status and revision therefore need to remain clear to the people working with them.

Engineering information that may require an explicit AI access decision includes:

• Requirements
• Risks and risk assessments
• Non conformances and related investigations
• Technical reviews and review evidence
• Test results
• Mission operations information
• Software and source code
• Security related technical information

Connecting AI to these sources changes how quickly information can be retrieved and combined. An AI connected to a document repository may search information across many documents within seconds.

This means organisations need to look beyond whether an engineer can open a document. They also need to decide whether an AI system may process that document as part of its task.

Understand what happens to engineering information in cloud AI

Cloud based AI services such as ChatGPT, Microsoft Copilot and DeepSeek give engineering teams access to capable language models without operating the underlying infrastructure themselves.

Before controlled engineering information enters an external AI service, the organisation needs a clear view of how that service handles the information. The answer can vary between providers, account types and technical configurations.

What happens to engineering information after you provide it to a cloud AI service?

Several areas deserve close attention.

• Retention and logging: prompts, uploaded documents and generated responses can be stored according to the service configuration. Teams need to understand where information is stored and how long it remains there.
• Model training and secondary use: organisations need to establish whether supplied information can be used for model improvement, service improvement or processing beyond the immediate engineering task.
• Information location: external AI services process engineering information within infrastructure outside the original engineering environment.
• Administrative control: account settings, contractual arrangements and technical configuration can influence how engineering information is handled.

Clear rules for external AI help engineers understand which information belongs in these services and which information stays within controlled engineering systems.

Separate human access from AI access

An engineer may have broad access because their role requires information from several disciplines. A systems engineer, for example, may work across requirements, interfaces and review evidence.

Giving an AI tool the same information access simply because it operates for that engineer creates a much wider AI information boundary. Many tasks only require a smaller set of sources.

Human access does not automatically need to become AI access.

AI access can instead follow the task. A requirements search assistant might receive access to approved requirements and selected supporting documents. A review assistant may receive a defined package of review evidence.

Define the information boundary before granting access

Cybersecurity and engineering teams can define the complete information path before connecting AI to engineering sources. This includes where information starts, where it is processed and where prompts or generated content may be stored.

Useful questions include:

1. What information will the AI be able to access?
2. Where will that information be processed?
3. Where can prompts, files and responses be stored?
4. How long can the service retain that information?
5. Can supplied information be used for model training?
6. Can access be logged and traced to a user, task or AI system?
7. Which sources are available to the AI for this specific use case?

The answers can become concrete access rules for each AI application. One system may work with public technical documentation. Another may process a selected set of controlled documents within organisational infrastructure.

Once the information boundary is clear, the next decision concerns what the AI may actually do with the information it receives.

Define what AI is allowed to do

Access gives AI information. Authority gives AI influence over engineering work.

An AI system might retrieve an approved requirement and present it to an engineer. It could also analyse that requirement, suggest a rewrite, prepare a change or interact with an engineering system. Each step gives the AI a different level of authority.

Ask yourself: once AI has access to the information, what is it allowed to do with it?

AI authority has a progression

A useful way to define AI authority it is to look at how close the AI moves towards changing controlled engineering information or influencing an engineering decision.

1. Retrieve: find and present existing information.
2. Analyse: compare information, identify relationships or prepare an assessment.
3. Propose: suggest wording, classifications, actions or engineering changes.
4. Modify: enter or change information within an engineering system.
5. Approve: accept engineering evidence, dispositions or formal decisions.
6. Execute: trigger an operational or technical action.

The engineering consequence can grow as AI moves through these levels. Retrieval keeps the original engineering record unchanged. Modification, approval and execution place AI much closer to engineering authority.

This concept also appears in NASA’s AI software assurance guidance, which calls for “documented authority boundaries” when defining human oversight of AI systems.

Ask yourself: do we want to compromise security, reliability, and safety by using a higher AI authority?

A convincing AI answer still requires engineering review

AI can produce technically fluent answers while working from incomplete context, an unsuitable source or an incorrect assumption. The impact becomes larger when the output moves closer to controlled engineering work.

Consider several examples:

• Risk assessment: AI can estimate probability or consequence and prepare a proposed risk classification. An engineer can then review that proposal against the programme method and available evidence.
• Requirement rewriting: AI can suggest clearer wording. The engineer can compare the proposal with the original intent, scope and verification method before accepting a change.
• Technical reviews: AI can search review evidence and prepare potential findings. The review team can assess interfaces, assumptions and evidence before formalising those findings.
• Non conformance assessment: AI can organise available information and prepare possible follow up actions. Engineering and quality roles remain responsible for the disposition.

These examples place AI in a supporting role while engineering authority stays with the people and processes responsible for the programme.

A search task can remain read only. Drafting can take place outside the controlled record. Formal engineering changes can continue through existing review and approval processes.

Bringing AI output into a controlled engineering process

For example, AI could prepare a rewritten requirement in a separate workspace. An engineer reviews the proposed wording against the original requirement and related engineering context. The accepted wording then enters the requirements management system through the established change process.

AI can prepare input for the engineering process. Engineers decide what becomes part of the controlled engineering record.

Match AI authority to engineering consequence

Giving AI access to engineering data is already an action with consequences. Even when an AI only reads information, that information leaves its original boundary and enters another processing environment.

If an AI can see your engineering data, you need to know what can happen to that data next.

Depending on the service and its configuration, information provided to AI may be processed, logged, retained or used to improve models. For controlled engineering information, read access therefore deserves its own security assessment. An AI does not need permission to change a requirement for its access to create an information security concern.

The question is not only what the AI can change. It is also what the AI service can do with the information it receives.

Lower impact AI activities can include:

• Searching approved engineering documentation
• Summarising selected documents
• Finding related requirements

These activities have a lower impact on the engineering record because the AI does not directly change controlled information. The data exposure still depends on where the AI runs and how the service processes the information it receives.

Higher impact AI activities can include:

• Changing data
• Accepting verification evidence
• Changing spacecraft configuration

These activities add engineering authority to information access. The AI can now influence controlled records, engineering decisions or spacecraft behaviour. This makes both the information boundary and the authority boundary relevant when assessing the use case.

Build AI around a reliable engineering environment

The engineering environment remains the reference for controlled programme information while AI models and services can change over time. The architecture can therefore place AI around established engineering controls instead of making the AI model the centre of the engineering process.

Ask yourself: how closely should AI connect to the controlled engineering environment?

Using local AI can offer greater control

Local AI can give organisations greater control over where engineering information is processed. Models can operate within infrastructure managed by the organisation, with connectivity and access configured around programme requirements.

The term “local AI” can describe several architectures. A model may operate on dedicated internal infrastructure or within another controlled computing environment. The appropriate setup depends on the engineering use case and the information involved.

Local deployment gives the organisation direct control over areas such as data location, connectivity, model versions and update processes. This can fit engineering environments where repeatability and controlled change matter.

The language model remains one component of the complete architecture. Document stores, retrieval systems, authentication, APIs, model servers and engineering integrations determine how information reaches the model and what happens with its output.

Where the model runs therefore answers only part of the question. The information it can access and the authority it receives remain separate design decisions.

Using cloud AI alongside the engineering environment

Space organisations can also use external AI services alongside their existing engineering systems. This gives teams access to current AI capabilities while allowing the organisation to set clear information boundaries.

The starting point is deciding which engineering information is suitable for the external environment. Public or specifically approved information can support one group of use cases. Controlled engineering information can remain inside the systems designed to manage it.

Generated content can move back into the engineering environment after engineering review. An engineer can compare the output with controlled sources and decide how it fits within the formal engineering process.

This makes the controlled engineering environment the stable reference point while AI services, models and capabilities continue to evolve.

Keep engineers in charge with ECLIPSE Software Suite

AI can help space engineers search technical information, connect engineering evidence and prepare analyses. Local AI can offer greater control over processing, while external AI services can support use cases built around approved information.

Across these architectures, the same boundaries apply. Define which information AI can see and what AI is allowed to do with it. Then design the system connection around the engineering consequence of the task.

ECLIPSE Software Suite provides an environment for managing controlled engineering information and structured processes throughout the space project lifecycle. Its solutions support areas such as document management, quality processes and risk management, giving teams a stable place to manage programme information as they introduce AI into their wider engineering landscape.

This creates a clear division of roles. AI can help engineers search, analyse and prepare information. ECLIPSE can remain the environment in which controlled programme information and formal engineering processes are managed. Engineers and established approval workflows determine what becomes part of the engineering record.

Explore ECLIPSE Software Suite solutions to see how a secure and reliable engineering environment can support your engineering information, or contact us for a free demonstration.