Does Kiro train on your code? It depends on how you sign in

The Morse teamPublished November 11, 202611 min read

As of November 2026, Kiro, the AI coding IDE and agent from AWS, does train on your code by default if you sign in with GitHub, Google, or AWS Builder ID, whether you use the Free tier or pay $20 to $200 a month. AWS's data protection page says it may use your questions, other inputs, "and the responses and code that Kiro generates" for service improvement, including "model training," unless you turn off content collection (Kiro). Content from users who sign in through AWS IAM Identity Center or a company identity provider, or with an Amazon Q Developer Pro subscription, isn't used for service improvement.

We read Kiro's data protection page (updated October 23, 2026), its FAQ, pricing, authentication, models, cloud sessions, memory, and billing docs, the AWS Service Terms (updated November 5, 2026), the AWS Privacy Notice, and Amazon Bedrock's abuse detection page on November 11, 2026. AWS writes much of this down. It names the exact checkbox in each of Kiro's three apps, publishes the 60-day abuse window, and says which AWS Region holds your prompts.

Kiro training and data collection: the key facts

Read on November 11, 2026, from Kiro's and AWS's own pages.

Does Kiro use your code to train models?

Yes, on the two kinds of account most individuals have. Kiro's data protection page sorts users into three groups: Free Tier users, "individual subscribers," and enterprise users (Kiro). For the first two, "Subject to your opt out, content may be used for service improvement." For enterprise users, "Content is not used for service improvement."

The content AWS names is the work itself: "your questions to Kiro, other inputs you provide, and the responses and code that Kiro generates." AWS lists four uses for it: to "provide better responses to common questions, fix Kiro operational issues, for de-bugging, or for model training."

The AWS Service Terms go further than the Kiro page. Section 50.3 lists "Kiro Free Tier, and Kiro individual subscribers" among the services where AWS "may use and store AI Content" to improve that service, may use AI content "that is not personal data to develop and improve AWS and affiliate machine-learning and artificial-intelligence technologies," and, for that improvement work, "may store such AI Content in an AWS region outside of the AWS region where you are using such AI Service" (AWS). So code collected through Kiro isn't limited to improving Kiro.

The AWS Privacy Notice, last updated May 18, 2026, doesn't settle this either way. It says it "does not apply to the 'content' processed, stored, or hosted by our customers," and sends you to the service agreement instead (AWS). For Kiro, that agreement is the Service Terms.

Does paying for Kiro Pro stop it training on your code?

No. The sign-in method decides it. Kiro's authentication page says "Users with a paid Kiro subscription who access it through a social login provider (like GitHub or Google) or through AWS Builder ID are considered individual subscribers," and individual subscribers are in the group whose content may be used (Kiro). As of November 11, 2026, Kiro's paid plans are Pro at $20 a month, Pro+ at $40, Pro Max at $100, and Power at $200 (Kiro). On any of them, a GitHub or Google login leaves content collection on until you turn it off.

Kiro's FAQ once promised the opposite. During the preview, it said: "For users who access Kiro with Pro or Pro+ tiers once they are available, your content is not used to train any underlying foundation models" (Kiro FAQ, archived July 17, 2025). CloudThat's Kiro guide, dated May 22, 2026, still tells readers "Your code is not used to train models" without naming a plan or a sign-in method (CloudThat). Kiro's FAQ today answers the same question differently: it excludes paid users "that access Kiro through AWS IAM Identity Center or external identity provider," and says AWS "may use certain content from Kiro Free Tier and Kiro subscribers that access Kiro through a social login provider or AWS Builder ID" (Kiro FAQ).

Kiro training by login: GitHub, Builder ID, or Identity Center

We built this from Kiro's data protection, authentication, FAQ, and pricing pages, read November 11, 2026.

The Identity Center route isn't open to a solo developer on the Free tier. Kiro's pricing FAQ says the Free tier "is not available in Kiro Enterprise or AWS GovCloud (US) Regions," which require a paid plan and enterprise sign-in (Kiro). Team credit pools start at $20 a seat a month for two or more seats.

For the same split at GitHub, where a paid organization rather than the sign-in method decides, see does GitHub Copilot train on your code.

How do you turn off Kiro content collection?

Uncheck it in every Kiro app you use. Kiro's data protection page gives separate steps for the IDE, the CLI, and Kiro Web, and each has its own switch (Kiro).

Kiro IDE:

1. Open Settings and switch to the User sub-tab.
2. Choose Application, then Telemetry and Content.
3. Uncheck Data Sharing and Prompt Logging: Content Collection for Service Improvement to stop content collection.
4. Uncheck Data Sharing and Prompt Logging: Usage Analytics And Performance Metrics to stop telemetry.

Kiro CLI: open Preferences, toggle off Share Kiro content with AWS, and toggle off Telemetry.

Kiro Web: sign in at app.kiro.dev, go to Settings, then Privacy & data, and toggle off Allow AWS to use your Kiro Web content for service improvement.

None of the pages we read says code collected before you opt out is removed, so change the setting before you open a repository you care about. For an AWS account managed through AWS Organizations, section 50.3 of the Service Terms adds a second route, an AI services opt-out policy set at the organization level.

What does Kiro telemetry collect?

Version, platform, and performance numbers, by Kiro's own list. The data protection page says telemetry covers "the Kiro version, operating system (Windows, Linux, or macOS), and the anonymous machine ID," plus the request count, errors, and latency for login, tab completion, code generation, steering, hooks, spec generation, tools, and MCP (Kiro). The IDE sends it to a telemetry endpoint in your profile's AWS Region, and drops it if that Region has no endpoint.

Telemetry and content collection are separate checkboxes. Turning off telemetry doesn't stop content collection, and turning off content collection doesn't stop telemetry. Separately, AWS says it "may collect operational signals" such as error logs and "metrics based on natural language instructions, feedback, and responses" to run the service, and the page gives no switch for those.

How long does Kiro keep your prompts and code?

Kiro publishes three periods, and none covers what it keeps for service improvement.

• Free tier, 60 days. "For Free Tier users we may store your inputs for up to 60 days" to detect abuse, and opting out of content collection "does not affect our ability to store your inputs for abuse detection purposes" (Kiro). AWS says it won't use those inputs to improve Kiro's models, but may use them "to develop and improve classifier tools designed to detect abuse."
• Some models, 30 days, for everyone. Kiro runs on Amazon Bedrock, whose abuse rules apply to all Kiro users. For OpenAI GPT models, "classifier-flagged traffic will be retained for up to 30 days." For Claude Fable 5.1, "all traffic will be retained for up to 30 days," and flagged traffic may get "human review performed by AWS." Bedrock's own page says retained inputs and outputs "are not shared with third-party model providers" (AWS).
• A deleted account, 30 days. The Kiro: Delete Account command starts a 30-day window to restore it, after which "your account and data will be permanently deleted" (Kiro).

Cloud sessions keep more. Kiro Web and Mobile run on them, and Kiro's docs say "conversation history, bound repositories, and the sandbox's file state persist in the cloud," with repositories cloned "server-side" from GitHub or GitLab (Kiro). The docs give no retention period. You remove a session with Delete Session or kiro-cli chat --delete-session.

Where does Kiro send your code?

To AWS, and from there to whichever model answers. Kiro's FAQ says the default, Auto, uses "a mix of different frontier models," and that you can pick OpenAI's GPT-5.6, Anthropic's Claude models, or open-weight models such as DeepSeek 3.2, MiniMax M2.5, GLM-5, and Qwen3 Coder Next (Kiro). Kiro's data protection page says it "is powered by Amazon Bedrock and uses cross-region inference."

If you're on the Free tier or an individual plan, Kiro stores your prompts and responses in US East (N. Virginia), and its models page says you use the US endpoint (Kiro). Most models stay within that geography. The three GPT-5.6 models have a Global inference scope, which means requests "may be processed in supported commercial AWS Regions worldwide" (Kiro). They route that way even from a Frankfurt profile.

Is Kiro safe for company code?

On an enterprise sign-in, AWS keeps your code out of training. Enterprise users "are automatically opted out of telemetry and content collection," their content isn't used for service improvement, and admins can encrypt data with their own AWS KMS keys (Kiro). All traffic uses TLS 1.2 or higher. Paid plans also carry AWS's output indemnity, which section 50.10 of the Service Terms extends to "Kiro (excluding Kiro Free Tier)," and not to "Kiro Excluded Models" such as Kimi K3 and anything marked experimental (Kiro).

Ask your admin whether prompt logging is on. It records "all inline suggestions and chat conversations" in the IDE and CLI, in an S3 bucket in the company's AWS account (Kiro). A personal GitHub login on a work laptop puts that same code back in the opted-in group, so sign in through your organization.

Asking about code on Morse, with no login type to pick

In Kiro, the answer turns on a login button. Your code reaches AWS under an account, and the account type decides whether the terms let AWS use and store it, where it may be stored, and which checkbox you have to find in which app. A $200 plan doesn't change that, and a missed checkbox in one of three apps leaves the code in the pool.

On Morse, no login type changes the answer. Morse doesn't train AI on your content on any plan, sells no data, and shows no ads, so there's no Content Collection for Service Improvement box to find. Morse is a chat app in your browser, where you can ask the code question you'd otherwise type into Kiro's chat, such as a failing test, a regex, or a function you want reviewed. The Free plan includes coding, file, and other chat tools. Pick a mode for each chat:

• Private, for a snippet you don't want sitting in a 60-day abuse log. Every new chat opens in Private. Open-weight models like DeepSeek, Kimi, and GLM, plus xAI's Grok, answer with your account, email, and IP address removed, and Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Here's what zero data retention covers.
• E2EE, for source code that no one but a verified enclave should see. Before anything leaves your device, your browser checks the enclave's hardware attestation, then encrypts the message for that enclave alone, and Morse's servers pass along ciphertext they can't read. The protection rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models with no web search or image generation, and it comes with Morse's Pro plan at $19 a month. How end-to-end encrypted AI chat works walks through each check.
• Anonymous, to ask Claude or GPT without your GitHub or Google identity attached. Morse's proxy strips your account, email, and IP address and sends the question under Morse's own account. The provider sees your code without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so swap real keys and hostnames for placeholders first.

Morse's Pro plan includes Claude Sonnet 5.5 and Grok 4.7, among other models. In every mode, your chat history is stored in your browser, encrypted, rather than on Morse's servers, and Morse's servers don't keep or log your messages. A file you attach goes with that message and Morse never stores it. Clearing your browser or deleting your account removes those chats for good, and Morse can't bring them back, so copy out any answer you'll need. The Morse privacy policy has the details, and which AI companies train on your data sets Morse beside the others.

What to change in Kiro before you open a repository

• GitHub, Google, or Builder ID, Free or paid: uncheck content collection in the IDE, the CLI, and Kiro Web separately, and turn off telemetry if you want that too.
• Free tier: expect your inputs to stay for up to 60 days for abuse checks whatever you set.
• Work code: sign in through your organization's IAM Identity Center, never a personal login, and ask whether prompt logging is on.
• A code question you don't want tied to your identity: ask it on Morse, in E2EE for source no one should read.

Try a chat from the Morse home page without an account, or start free with no card.