Note: This information was compiled from details provided by Anthropic as of 28.08.2026. We will continue to monitor for any changes and update this help note should there be significant developments.
This article sets out, in detail, what happens to your data when someone generates an AI Report, what's sent to Anthropic, how and where it's processed, and what does and doesn't happen when a report or its underlying session is deleted. It's intended for IT, security and privacy teams assessing the feature before enabling it.
What gets sent to Anthropic
Three things are sent when a report is generated:
The request itself, the report name, description, your instructions, the date range and the output format.
The project records the assistant retrieves in order to write it, stakeholders, interactions, complaints, commitments, tasks, form responses, outgoing communications and project properties, including any free-text notes and personal details they contain.
The finished Word and PDF file, which is produced on Anthropic's side and then copied back to us.
Anthropic cannot use this connection to reach other projects or information outside of what that specific report is permitted to access.
Who processes it
Anthropic PBC, under their commercial terms for business customers, not the terms that apply to personal Claude accounts. The model is Claude Sonnet 5, running on Anthropic's Managed Agents platform.
Each of our regions has its own isolated Anthropic workspace.
The assistant's sandbox can only reach one destination, our own data endpoint, and cannot install software.
Its credential is a 10-minute token, scoped to a single report run.
Access boundaries
A report can only ever contain what the person requesting it could already open in Simply Stakeholders. Project scope, confidentiality levels, organisation and date window are all enforced by our server, not by the AI. Asking the assistant for more doesn't widen what it gets back.
Training
Anthropic's published commitment is that customer data is never used for model training without express permission. We haven't given that permission, and won't. Nothing from your data becomes part of a Claude model, and it can't appear in another organisation's report.
Where it's stored
Anthropic currently only offers US-based storage for this platform, so stored data sits in the United States regardless of whether you're on our EU, UK, Australian or Canadian instance. Our per-region workspaces separate one region's data from another's, they don't move it out of the US.
Model processing is also not restricted to a single territory by default. Anthropic offers a US-only processing option and we can enable it on request, but it isn't on by default and doesn't follow automatically from your hosting region.
How long it's kept
On our side, generated report files (Word, PDF and preview images) are kept in our storage indefinitely, until the report is deleted or a file is manually removed.
On Anthropic's side, two different things are retained differently:
The checkpointed working files from the run, including the copies of the generated Word and PDF made while the report was being produced, fall under Anthropic's own 30-day platform limit for sandbox state.
The session history, every prompt, tool call, tool result and AI response, i.e. the actual data sent to generate the report, is not covered by that 30-day limit. Anthropic's documented position (their Managed Agents session and event-stream documentation) is that this persists until explicitly deleted. We don't yet run an automatic job to delete it, so for now this should be treated as retained indefinitely at Anthropic.
Separately, and for all Anthropic customers: if their automated safety systems flag a session, or a legal obligation applies, they may retain inputs and outputs for up to two years. That can't be configured away, by us or by you.
Deleting a report
Deleting a report in Simply Stakeholders removes it from the list and withdraws access immediately. A scheduled cleanup then deletes the report's Word, PDF and preview files from our storage, and confirms that Anthropic's copies of those same files have also been removed, only once that's confirmed do we close out the Anthropic session. If that check fails, we retry later rather than closing the session too early.
Closing the Anthropic session archives it rather than deleting it. The underlying prompts, tool calls, tool results and AI responses are retained by Anthropic for audit purposes, and aren't removed by deleting the report. If you need that underlying session data removed as well, this can be requested through support and actioned manually, including a bulk request covering all AI Report data for your organisation.
If you're doing a risk assessment: real stakeholder data leaves our environment to generate a report, and is stored and processed in the United States. It is not used to train Anthropic's models. The generated files are cleaned up when you delete a report, but the underlying Anthropic session, which contains the data sent to create it, is retained for audit purposes and isn't automatically purged today. Existing Simply Stakeholders permissions still apply, a report can never contain more than the requester could already see. The feature is optional and can be left switched off for your organisation. If you have data residency obligations, a restriction on data leaving your jurisdiction, or a defined retention limit for stakeholder records, raise it with your Customer Success Account Manager before enabling it.
For more help
Contact our Customer Success team via email [email protected].