1. Our commitment
We take the security and safety of our systems and our users seriously, and we value the work of researchers who help us keep them safe. This policy explains what you may test, how to report a problem, what you can expect from us, and how we protect good-faith researchers.
2. How to report
Email security@openwist.com. We also publish this contact at openwist.com/.well-known/security.txt.
Please include:
- a clear description of the issue and its impact;
- the affected system, address, product, version or model;
- the steps needed to reproduce it, a proof of concept, or screenshots;
- any ideas on how to fix it; and
- whether you would like to be credited, and how.
Do not include personal data about other people in your report. If you need to share sensitive material, ask us and we will arrange a secure way to do so.
3. Scope
This policy covers systems and software that Openwist owns or operates, including:
- our websites and administration systems;
- Exa on the web, on messaging apps and through any other interface;
- the services and APIs that run our products;
- tools and packages we publish, including our repositories and extensions;
- models we publish; and
- software for vehicles and robots that we release.
We are especially interested in:
- flaws in authentication or access control, including access to another user's data;
- injection and scripting flaws, including through generated output or shared content;
- ways to escape an isolated code environment or reach internal systems;
- exposure of secrets, credentials or personal data;
- ways to bypass usage limits or payment controls at scale;
- supply-chain weaknesses in anything we publish; and
- any flaw that could affect the physical safety of people, vehicles or machines. We treat these as the highest priority.
4. Out of scope
- Services we do not own, such as cloud platforms, messaging platforms, payment processors, package registries and app stores. Please report those to their owners.
- Denial-of-service attacks, volumetric tests, spam and anything that degrades the service for others.
- Social engineering or phishing of our staff, contractors or users, and physical attacks on our premises or equipment.
- Output from automated scanners with no demonstrated impact.
- Missing hardening settings or headers without a way to exploit them.
- Issues that need an already-compromised device or that affect only your own account.
- Reports about outdated or unsupported browsers and platforms.
- Everyday inaccurate answers from an assistant with no safety, security or privacy consequence. Use the feedback control in the product for those.
5. Rules for testing
To be covered by the safe harbour in section 8, you must:
- Use your own accounts and data. Do not access, change, copy or delete data that is not yours. If you reach another person's data by accident, stop, do not keep or share it, and tell us.
- Do the minimum necessary to demonstrate the issue, and do not go further into our systems than needed.
- Keep volume low. Do not run load tests, mass automated requests or anything that disrupts other users.
- Do not test on real users, or on messaging channels with people other than yourself.
- Do not test vehicles or robots on public roads, near people or in any way that could cause harm. For physical systems, work only on equipment we have provided to you under a written agreement, or on simulations.
- Keep the details private until we have had a reasonable time to fix the issue (section 7).
- Do not break the law beyond what a good-faith test of these systems unavoidably touches.
6. What you can expect from us
| Step | Our target |
|---|---|
| Acknowledge your report | Within 5 business days |
| Confirm the issue and give an initial severity rating | Within 10 business days |
| Fix or mitigate | By severity: critical issues as fast as we can, aiming for 7 days; high within 30 days; medium and low within 90 days or the next scheduled release |
| Keep you updated | At least every 30 days until it is resolved |
| Tell you when it is fixed | On release, and credit you if you wish |
We rate severity by the impact on people and data, how easy the issue is to exploit, and standard scoring such as CVSS, with extra weight on anything that affects physical safety. These are targets. If we cannot meet one, we will tell you why.
7. Coordinated disclosure
We ask you to give us 90 days from our acknowledgement before you publish details, or until a fix is released if sooner. If we agree a fix needs longer, we will say so and agree a new date with you. If an issue is already being exploited or puts people at serious risk, we may ask to move faster, and we may ask for longer where a physical product must be updated in the field.
We will not ask you to stay silent indefinitely. If we cannot fix a valid issue, we will work with you on a responsible way to warn the people affected.
8. Safe harbour
If you act in good faith and follow this policy:
- we consider your research to be authorised;
- we will not take legal action against you, or ask law enforcement to do so, for activity that follows this policy; and
- if a third party takes action against you for research that followed this policy, we will state that your research was authorised.
This policy covers only systems that Openwist owns. We cannot authorise testing of other organisations' services, and we cannot bind courts or authorities. If you are unsure whether something is allowed, ask us at security@openwist.com before you test.
9. Recognition
We do not currently offer monetary rewards. With your permission, we will thank you by name or handle on an acknowledgements page once the issue is fixed. We may introduce a rewards programme in the future and will update this policy if we do.
10. Reporting AI safety problems
Some problems are not conventional security flaws, such as a way to make an assistant produce seriously harmful content, to disable its safeguards, or to cause unsafe behaviour in a system. Send these to safety@openwist.com with the exact prompts or steps and the output you received so that we can reproduce them. We handle them like security reports. Because model behaviour is not always repeatable, we may ask you to help test a fix. Please do not publish a working method for causing serious harm before we have had time to respond.
11. Unsafe products
If you believe a vehicle or robot supplied by Openwist is unsafe, stop using it and contact recall@openwist.com. If anyone is in immediate danger, call your local emergency services first. Product defects are handled under our Accountability Terms and Safety Framework, not as security reports.
12. Your data
We handle the contact details and report content you send us under our Privacy Policy. We use them to investigate, to communicate with you, and to credit you if you ask. We keep them for as long as needed for these purposes and for our records.
13. Changes
We may update this policy. The current version is always published on our website, and earlier versions are kept.
Contact: security@openwist.com · AI safety: safety@openwist.com