Security we can show you.
Not security we only claim.
What Opscale does to protect client data, written down: a signed NDA before any information is shared, Answer acting only inside the rules you set, a hardened public site, and an outside-in check by our security partner, ZetTateK. Nothing on this page is a promise we cannot evidence.
Security partner
01
Four things in place
in every engagement.
An NDA before any data
A non-disclosure agreement and a master services agreement are signed before any client information is shared. Both are yours to read before you sign.
Answer acts only inside the rules you set
Answer books on its own only inside your rules: your open slots, the job types you take and your emergency rule. Anything outside them is drafted for you and waits, and the approval tap is the send.
A hardened public site
hireopscale.com runs behind strict transport security, a content security policy and the standard browser protections, and publishes a security.txt contact so researchers can reach us.
An outside-in check by our partner
ZetTateK, our security partner, ran an outside-in exposure check of our domain in August 2026. Every finding was reviewed and the record is on file.
02
What security looks like
in a real engagement.
Five checkpoints, from the first agreement to your security questionnaire. Each one is something you can check yourself.
Agreements before any information
A non-disclosure agreement and a master services agreement are signed first. Nothing about your business is shared with anyone until they are.
Access granted by you, in your own systems
The work runs on access you grant inside your own accounts. You can see it, scope it to what the job needs and revoke it at any time.
Your rules first, then your approval
Answer acts only inside the rules you set. Emails, messages and documents outside them are drafted for you, and the approval tap is the send.
Every approval and every send recorded
Each outgoing message and each approval is logged inside your instance, so what went out and when is never a matter of memory.
Your questionnaire answered in writing
Send us your security questionnaire. Every item gets what is in place today and what is in progress, with dates. Nothing is answered with a template.
03
The controls,
each one on the record.
Every line here is an answer in our written security questionnaire, with the evidence behind it. Ask for the questionnaire and you can check each one.
Who can get in
Only named logins reach your workspace
Your workspace on the platform is reached only through Cloudflare Access, with a Google login on a list of named people, checked before any request reaches the server.
Roles, not shared power
Built into the platform behind Answer, Close and Collect: owner, operator and viewer roles, each with its own list of what the AI may do. A tool a role may not use is never offered to the AI.
A server with one locked door
Root login and password login are off. The only open port is a rate-limited admin login; web traffic arrives through an outbound encrypted tunnel.
What the AI can do
A tap bound to a person
Every send, approval or fix the AI proposes outside your rules waits for a tap. The tap carries a single-use token bound to the approver’s login and role.
Mail is data, not orders
Text from email, documents and the web reaches the model labelled as untrusted. Once it has been read, every action with a side effect pauses until the owner clears it.
A log that shows tampering
Built into the platform behind Answer, Close and Collect: every action recorded with who, what, from where, when and the outcome, in a hash-chained log written by a separate system user the app cannot impersonate.
Where your data goes
Hosted in the United States
A new client workspace is deployed on a server in DigitalOcean’s New York region. Every provider that may receive data is listed on our subprocessor page, with what it receives.
Not used to train models
Anthropic, the model behind the agents, may not train models on customer content under its commercial terms.
Encrypted in transit
TLS 1.3 on the website and on every platform address, with older protocols refused and strict transport security on.
Secrets out of code, dependencies checked
A scanner refuses any change that carries a secret, personal data is kept out of our code repository, and dependencies are audited every week.
The full list of who receives what is on our subprocessor page. How we handle personal data is in the privacy policy.
04
If something
goes wrong.
Notice within 72 hours
If a security incident affects your data, you hear from us within 72 hours of confirming it. The master services agreement says so.
A written plan and an honest register
Four severity levels, who is called and when, then contain, preserve, assess, notify, fix. Every incident goes into a register with its cause and its fix, our own mistakes included.
Report a vulnerability
Write to jose@hireopscale.com, in English or Spanish. The contact is also published in our security.txt.
Yours on request, under NDA
- Our answers to a standard security questionnaire, each one with its evidence
- The information security policy and the incident response plan
- The disaster recovery plan and its restore drill record
- The subprocessor list, which is also public
An insurance certificate and an independent penetration test letter join this list when they exist, not before.
05
What we do not claim.
What is in place and what is in progress, with dates. Ask for it on the call.
No certification we have not earned. Opscale holds no SOC 2 or ISO 27001 certificate, and says so.
No questionnaire answered with a template. Each item gets what is true today, with a date.
No service attributed to a partner that the partner does not deliver for you.
No number on this page we cannot show you the record for.