Amazon Q Developer launched at AWS re:Invent in November 2023, replacing CodeWhisperer as Amazon's primary AI coding tool. It's free at the base tier and $19/user/month for Pro. If you're building on AWS, you've probably tried it or been told to. Here's the honest security picture, starting with the part that changed everything.
Amazon Q Developer is end-of-life. AWS announced on April 30, 2026 that Q Developer IDE plugins and paid subscriptions reach end of support on April 30, 2027. New signups were blocked from May 15, 2026, so if you don't already have a subscription you can't start one. AWS points existing users at Kiro and publishes a migration guide. Everything below still describes how Q behaves today, but treat it as a wind-down, not an adoption decision.
TL;DR
Amazon Q Developer is end-of-life: new signups stopped May 15, 2026 and IDE plugins lose support April 30, 2027. While it runs, it's one of the safer AI coding tools. The Pro tier never uses your code for training, built-in code review catches OWASP Top 10 issues, and it runs on AWS's own audited infrastructure. The free tier is also reasonably private since Amazon updated its data policy in late 2023. The risk isn't Q itself. It's the code Q generates. Always scan before shipping.
What is Amazon Q Developer?
Amazon Q Developer is AWS's AI coding assistant. It replaced CodeWhisperer in 2023 and added agentic capabilities: it can write multi-file changes, fix bugs across a codebase, and generate unit tests automatically. It integrates with VS Code, JetBrains IDEs, the AWS Console, and AWS Cloud9.
Two separate questions matter here:
- Is the tool itself secure? Where does your code go, who sees it, is it used for training?
- Is the code it generates safe? Can you ship what Q produces without a security review?
The answer to question 1 is mostly yes. The answer to question 2 is: not without checking.
Our Verdict
What's Good
- Pro tier never trains on your code
- Built-in OWASP code review (no extra tool needed)
- Runs on AWS's own infrastructure (SOC 2 Type II, ISO 27001, FedRAMP High)
- Strong IAM-based enterprise controls
- Free tier with updated privacy policy since 2023
What to Watch
- End of support April 30, 2027, and closed to new signups since May 15, 2026
- Code is still sent to AWS servers for every completion
- AI-generated AWS SDK code can expose IAM credentials if accepted uncritically
- Security scanning doesn't catch business logic flaws
- Amazon Q agent features send broader code context than basic autocomplete
Privacy and Data Handling
This is the part that matters most to most teams. Here's the actual breakdown:
| Free | Pro ($19/user/month) | |
|---|---|---|
| Code used for training | No (opt-in only, since 2023) | Never |
| Code sent to AWS for completions | Yes | Yes |
| Admin controls | None | Full |
| SSO / SCIM | No | Yes |
| Code review quotas | Size-based, not a monthly count | Size-based, larger project reviews |
| Agentic requests/month | 50 | Expanded |
| New signups | Closed since May 15, 2026 | Closed since May 15, 2026 |
Amazon updated the free tier policy in late 2023. Before that, code was used for model training by default. Now it isn't, unless you explicitly opt in. The paid tier has always had a hard "never" on training.
The tier names moved with the rename. CodeWhisperer's Individual and Professional became Q Developer's Free and Pro, and AWS notes the Pro admin setup is genuinely different from CodeWhisperer Professional rather than a relabel. Guides still using the old names predate April 2024.
What does "code sent to AWS servers" mean in practice? When Q generates a suggestion, it sends your open files and cursor context to Amazon's backend (running on Amazon Bedrock). That data is processed in-memory for the completion and not stored long-term. For most apps, this is fine. For code with hardcoded secrets, it's a problem (though Q's secret detection should flag those before they reach the backend anyway).
Agentic mode sends more. When you use Q's agent features (multi-file edits, /dev tasks), it reads a broader slice of your codebase to plan changes. If your repo contains .env files, AWS credentials in config files, or other secrets, make sure those are in .gitignore and excluded from Q's workspace scan.
Built-in Security Scanning
Amazon Q's security scanning is its biggest differentiator. It's powered by Amazon CodeGuru Security, which runs over 1,100 detectors from the CWE (Common Weakness Enumeration) database.
What it catches:
- Hardcoded credentials and API keys (including AWS access keys)
- SQL injection
- Cross-site scripting (XSS)
- Path traversal
- Insecure cryptography (MD5, SHA-1 for passwords)
- OWASP Top 10 vulnerabilities
What it misses:
- Business logic flaws
- Context-specific authorization gaps (e.g., missing ownership checks)
- Infrastructure misconfigurations (use AWS Security Hub for those)
- Runtime vulnerabilities that only appear under specific conditions
Quotas are measured by size, not by a monthly count. AWS documents auto reviews capping at 200 KB of input artifact and source, while file or project reviews allow up to 500 MB of input artifact and 50 MB of source. Older guides quoting "50 scans per month" for free and "500" for paid are describing the retired CodeWhisperer tiering.
One behaviour matters more than the limits: by default Q reviews the output of git diff on the active file, not your whole project. Code committed before you installed it was never examined unless you explicitly ask for a project review.
Run a project-wide scan before deploying. Q's file-by-file scanning during development catches issues as you write. The full project scan before shipping catches patterns that only appear when you look at the whole codebase, like an API route that lacks auth because the middleware that protects it was defined in a file Q hadn't seen.
The Real Risk: AI-Generated AWS Code
Amazon Q excels at AWS-specific code: Lambda functions, IAM policies, CloudFormation templates, DynamoDB queries. That's also where the biggest security gaps tend to appear.
CheckYourVibe's scanner finds AWS access keys (pattern: AKIA[0-9A-Z]{16}) in frontend JavaScript in roughly 12% of AI-built apps that use AWS SDK patterns. The typical cause: the developer asked Q to "add S3 upload functionality," Q generated working code using hardcoded credentials, and the developer shipped it without swapping in environment variables.
This isn't Q's fault. It's a pattern-completion tool that produces what seems most likely based on context. If your codebase has credentials scattered in config files, Q's suggestions will reflect that. The fix is the same regardless of which AI tool you use:
Never accept Q's credential suggestions literally. When Q suggests new S3Client({ accessKeyId: "AKIA...", secretAccessKey: "..." }), that's a placeholder pattern. Replace with process.env.AWS_ACCESS_KEY_ID before committing.
Use IAM roles, not access keys, for services running on AWS. Lambda, EC2, ECS, and Elastic Beanstalk can all assume IAM roles. Q knows this and will suggest it if you prompt correctly: "write S3 upload code that uses IAM role-based auth, no hardcoded credentials."
Run CheckYourVibe before shipping. Q's built-in scan catches many issues, but it's checking within the IDE context. An independent scan after you've integrated everything catches patterns that span files.
Amazon Q vs Competitors
| Amazon Q Developer | GitHub Copilot | Cursor | |
|---|---|---|---|
| Built-in security scanning | Yes (1,100+ detectors) | Yes (Copilot code review) | No |
| Training data opt-out | Free: yes; Pro: never | Requires Business/Enterprise | Privacy mode available |
| Enterprise admin controls | Strong (IAM) | Good (GitHub Org) | Basic |
| AWS-specific suggestions | Excellent | Good | Good |
| Free tier | Yes | No | Limited |
| Agentic features | Yes (/dev tasks) | Yes (Copilot Workspace) | Yes |
If your stack is AWS-heavy and you don't run separate SAST tooling, Amazon Q's built-in scanner gives it a clear edge. If you're on a mixed stack or GitHub-native workflow, Copilot's ecosystem integration may matter more.
Compliance and Certifications
Amazon Q Developer inherits AWS's compliance posture:
- SOC 2 Type II: security, availability, confidentiality
- ISO 27001: information security management
- FedRAMP High (Q runs on Bedrock, which holds this authorization)
- HIPAA eligible for Pro tier deployments
- GDPR: AWS Data Processing Agreement available
For most startups, these certifications aren't the day-to-day concern. They matter when enterprise customers or regulated industries do vendor security reviews.
Is Amazon Q Developer safe to use?
Safe to use, but it's closing down. AWS blocked new signups on May 15, 2026 and ends support for the IDE plugins on April 30, 2027. On privacy: the Pro tier ($19/user/month) never uses your code for training and gives you admin controls, while the free tier has improved privacy since 2023 but still sends code context to AWS servers for completions. The bigger risk is the code Q generates. Always scan it before deploying.
Does Amazon Q use my code to train its AI?
For the Pro tier: no. Amazon explicitly states that customer code is never used to train foundation models. For the free tier: Amazon updated its policy in late 2023 so your code is not used for training unless you opt in. Note the tier names changed with the 2024 CodeWhisperer rename, so older guides say Professional and Individual for these same two plans.
What security vulnerabilities does Amazon Q detect?
Amazon Q's built-in code review detects OWASP Top 10 issues, hardcoded credentials, SQL injection, XSS, insecure cryptography, and over 1,100 patterns from the CWE database, plus infrastructure-as-code misconfiguration and third-party dependency issues. Quotas are size-based (200 KB for auto reviews, up to 50 MB of source for project reviews), not the monthly scan counts older guides quote. By default it reviews your git diff, not the whole project.
Is Amazon Q better than GitHub Copilot for security?
That gap has closed. Q's built-in scanner used to be the standard argument for it, but GitHub's documentation now says Copilot code review gives feedback on bugs and security vulnerabilities too, with a Balanced mode for security-sensitive code. For AWS-heavy projects (Lambda, CloudFormation, IAM policies), Q's suggestions tend to be more accurate. Copilot edges ahead on general-purpose languages and IDE integration depth, and it isn't scheduled for end of support in 2027.
Can Amazon Q generate insecure code?
Yes. Like any AI coding tool, Q can produce code with hardcoded secrets, missing input validation, or flawed auth patterns. CheckYourVibe's scanner finds AWS access keys (AKIA... prefix) in frontend JavaScript in roughly 12% of AI-built apps that use AWS SDK patterns , often from blindly accepting AI suggestions. Run a scan before you ship.
Using Amazon Q to Build Your App?
Scan your project for issues Q's scanner misses: exposed AWS credentials in frontend code, missing auth on API routes, and business logic gaps.