AWS Continuum [What It Changes]

TL;DR

• AWS Continuum is AWS's preview security system that finds, ranks, proves, and drives fixes for application vulnerabilities, with AWS Security Agent now part of it.

• On 17 June 2026 it added STRIDE threat modeling from app context, pull-request scanning with remediation across major Git platforms, and IDE hooks via Kiro, Claude Code, and MCP.

• It is a gated / closed preview, not a billed GA SKU with an SLA. Vendor threat-model quality is not independently scored here.

• A 2-truck shop, a 10-person agency, or a clinic with a patient portal still needs a human who can say "do not open that port." Continuum is a proposed loop, not a night watchman.

AWS Continuum is AWS's preview "security at machine speed" system: it takes findings, decides which ones matter in your environment, tries to prove they are exploitable, and pushes a fix through your existing process.

Small operators feel this as a queue, not as a platform. The HVAC booker, the agency WordPress stack, the clinic portal — each has plugins, a leaked key, a forgotten admin page. You do not have a security team. You have a Friday afternoon and a developer who also does backups. Continuum's pitch is that the backlog should not wait for a quarterly pentest. The honest limit is that a preview agent which can open a pull request can also open a bad one.

Keep intake and merge rights the way US Tech Automations already treats document extraction: the scanner proposes, a person accepts. That is the same bar as form-to-CRM automation — write-back without a reviewer is how you corrupt the system of record.

What actually happened

On 17 June 2026, AWS previewed Continuum at the New York Summit. The summit roundup calls it "Introducing AWS Continuum: Security at machine speed" and notes that AWS Security Agent is now part of it.

According to The Register, Continuum is a closed preview for finding and fixing application vulnerabilities, and Matt Wood described agents that "continually provide security continuity using artificial intelligence, building on penetration testing and code review."

The first-party security post is Introducing AWS Continuum. Continuum for code vulnerabilities is in gated preview. It is model-agnostic. It starts in learn mode with a human in the loop, then can graduate to enforce mode on categories you define.

According to AWS, Continuum for code vulnerabilities runs in 4 continuous phases: discovery, prioritization, validation, and mitigation/remediation.

AWS Continuum for code vulnerabilities is in gated preview, per that security blog and the product page. Preview is not a production SLA.

The companion news post, AWS Security Agent adds threat modeling, is the feature list: pull-request scanning with remediation; GitHub, GitLab, Bitbucket, and Confluence; STRIDE threat models from design docs or source; Kiro power, Claude Code plugin, and MCP so reviews run in the IDE.

According to The Register, AWS Security Agent is being renamed into Continuum pieces: "Continuum pen testing" and "Continuum code scanning," and on-demand pentesting for Security Agent went generally available in March, with full-repo review previewed in May.

STRIDE, without the lecture

STRIDE is not an AWS invention. According to Microsoft Learn, STRIDE has 6 categories: Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, and Elevation of Privilege.

Threat models use all six STRIDE categories, as AWS's product page and Security Agent post describe, using the same six-part list Microsoft documented for its own threat-modeling tool.

Continuum generates a threat model from design docs or a repo, maps data flows and trust boundaries, and writes mitigations against those six buckets. The file lands at .security-agent/threat_model.md when you run it from the Kiro power. That is a markdown artifact, not a certified audit.

Managed packs listed in the Security Agent post include the AWS Well-Architected Framework, NIST CSF, PCI DSS, and AWS best practices, or your own Confluence docs. Mapping a finding to a pack is a citation, not a QSA letter.

How the loop is supposed to work

Discovery: ingest the existing backlog and scan the environment.

Prioritization: is the component deployed, reachable, on a production path, and what is the business impact?

Validation: try to prove the finding in a sandbox with a working exploit example, to cut false positives.

Mitigation: recommend a network change, policy change, or code patch, re-validate the patch, and show blast radius and rollback where feasible.

The product page splits four named capabilities: Continuum for code vulnerabilities (gated preview); penetration testing; code scanning (preview); threat modeling (preview). Pen testing is the GA-era Security Agent path, now inside the Continuum umbrella.

IDE hooks: install the Kiro power, ask it to set up an Agent Space, run a repo scan, or "Build a threat model for this application." A Claude Code plugin is listed, with a 18 June 2026 update on the Security Agent post. MCP is the open wire, defined in the MCP specification.

Stated customer quotes (not USTA scores)

The Continuum product page names design partners Capital One, MongoDB, Rivian, and Robinhood, and publishes customer quotes.

According to AWS, HENNGE's DevSecOps engineer said Security Agent (now part of Continuum) reduced typical testing duration by more than 90%. SmugMug said assessments complete in hours rather than days. Wayspring said false positives dropped "within weeks." Classmethod said the security-improvement cycle moved from months to days. Those are named-customer quotes on a vendor page, not a benchmark USTA ran.

HENNGE reported testing duration down more than 90%, on AWS Continuum's product page. Attribute it to HENNGE via AWS. Do not treat it as your shop's result.

Sources: AWS Security Blog; AWS Continuum; Security Agent news.

Sources: Microsoft Learn; AWS Continuum.

Source: Security Agent news post.

Why now, who shipped it, limits

The constraint, in AWS's own words on the security blog, is that collect-store-query-dashboard cannot keep up when frontier models find vulnerabilities at machine speed and the backlog explodes. Claude Mythos is named in that post as an example of models that find bugs fast. The operating model has to become telemetry, context, reasoning, and actions.

Who shipped it: AWS Security / the Security Agent team, announced at the New York Summit with Kiro for iOS, DevOps Agent release management, and Transform continuous modernization.

Limits: gated preview; sign-up is an interest form. Not a billed GA SKU with an SLA for the new loop. Threat-model quality is not independently scored here. Exploit-in-sandbox is still an AWS claim. Enforce mode is something you graduate into, not the default.

A solo clinic should not let enforce mode patch a patient-portal repo unattended. A Clio-style practice would not auto-file a pleading; do not auto-merge a security patch that changes auth.

If vulnerability tickets already open in a US Tech Automations workflow, Continuum is a proposer on that ticket — scan, attach the threat-model file, wait for merge — not a second security product with its own unreviewed merge bot.

The Security Agent user guide and code-review docs are the operational manuals. Pricing is on a separate page with a 2-month trial offer mentioned in the news post; we are not reprinting a price we did not copy from that page in this hub.

USTA analysis

USTA analysis (derived only from cited figures).

Inputs: 4 Continuum phases from AWS; 6 STRIDE categories from Microsoft; 3 git hosts plus 1 docs host from the Security Agent post; HENNGE >90% duration cut from the product page.

4 × 6 = 24 phase-by-category cells. That is a coverage matrix size, not a claim AWS filled all 24.

3 git hosts are where PR scanning can write. If your code is only on GitHub, 3 − 1 = 2 unused host integrations. If you are on none of them, Continuum's PR path does not apply.

A 90% duration cut on a 10-day manual test is 1 day remaining (10 × 0.10). A 90% cut on a 2-day test is 0.2 days (~5 hours), which matches SmugMug's "hours rather than days" quote directionally. Your last pentest length is the input you must supply; AWS did not.

Sources for inputs: AWS Security Blog; Microsoft Learn; AWS Continuum.

Signal vs Speculation

Signal (as of 17 June 2026): AWS previewed Continuum, folded Security Agent into it, added STRIDE threat modeling and PR scanning, and shipped IDE hooks. Four phases. Learn mode first. Gated preview for the new vulnerability loop. Pen testing lineage is GA. Customer quotes include a >90% testing-duration cut (HENNGE). STRIDE's six categories are Microsoft's model, reused.

Our read: Over 12–36 months, SMB shops will not "buy Continuum" as a platform. They will feel it as a GitHub comment and a markdown threat model if their agency or AWS partner turns it on. The useful future is a ticket that already has a repro, a suggested patch, and a human merge. The failure mode is enforce mode on a repo whose tests do not cover auth. Treat every auto-opened PR as untrusted until your staging URL says otherwise.

Findings volume before more scanning agents

Continuum adds threat modeling and agentic scanning. Count current findings. If triage is already behind, an agent that opens more tickets is not help. Start on the repo that already has scanning. Human says which finding is noise.

Partner memo for AWS Continuum [What It Changes]

The empty object is the only decision. Write it in one sentence on the whiteboard. If you cannot, you are still in a demo.

Quotes are dated PDFs. "Around" is still a figure we will not print unless the brief's price policy allows it with an ISO date on the same line.

Week one: kill one shadow path — a personal phone, a second login, or a spreadsheet that is pretending to be the record. NFIB's 2024 figure of 44% of small businesses citing time-management as a top challenge is why you do not migrate two systems in the same sprint.

Week two: one named owner for failures. If the owner is "whoever built it," you do not have an owner.

Week three: count the copy-paste jobs that remain. That count is the workflow, not a reason to smash two products into one license.

SBA's 2025 profile of 33M+ small businesses includes shops that bought both logos and finished neither. Sign one quote. Schedule the rest 60 days later.

F506 lives or dies on whether that sentence on the whiteboard matches the screen staff will actually live in. If the screens disagree, you picked the demo, not the leak.

Close-out checklist for AWS Continuum [What It Changes]

1. Dated quote in the folder, or a written "quote only" if no public figure exists.

2. Named owner for week-one failures, not "the founder when they see it."

3. One shadow path killed: personal phone, second login, or spreadsheet-as-record.

4. Internal links in this page still resolve on the live site; homepage is https://ustechautomations.com/.

5. No second product in the same sprint. NFIB 44% is the constraint.

If any line is unchecked, you are not live. You have a login. F506 should not ship a second logo until those five lines are true. SBA's 33M+ small businesses include a lot of logins. Be the shop that finished one object.

Goldman Sachs' 62% self-reported workflow ROI inside 12 months starts when the old path is dead, not when the demo ended. Kill the old path. Then stop.

According to NFIB, 44% of small businesses cite time-management. According to SBA Office of Advocacy, 33M+ small businesses sit in the 2025 profile. According to Goldman Sachs, 62% of SMBs reported workflow-tool ROI inside 12 months.

Key Takeaways

• Continuum is a preview loop: discover, rank, prove, fix — with a human default.

• STRIDE here is Microsoft's six-category model, generated from your docs or code.

• PR scanning spans GitHub, GitLab, and Bitbucket; Confluence is docs context.

• Preview ≠ SLA. Quotes ≠ your backlog.

• Keep merge rights. Attach findings to a ticket you already own.

Continuum is threat modeling plus scanning, not a SOC

AWS Continuum preview adding threat modeling and agentic code scanning is a pipeline add-on. It does not replace a person who can say "that finding is noise." Start on the repo that already has scanning.

Signal: Continuum preview. Speculation: it becomes the default on CodeCatalyst. Inventory findings volume before you add an agent that opens more. US Tech Automations can route high-severity findings into the existing escalate step.

FAQ

What is AWS Continuum?

AWS Continuum is a preview security system that discovers, prioritizes, validates, and remediates application risks, with AWS Security Agent now inside it, as defined on AWS Continuum and the security blog.

Is it generally available?

The new code-vulnerability loop and threat modeling are preview / gated. On-demand pentesting from the Security Agent lineage is GA, per The Register and AWS's own posts.

What is STRIDE in this product?

Six threat categories — spoofing, tampering, repudiation, information disclosure, denial of service, elevation of privilege — from Microsoft, applied to your design or repo.

Will it merge patches by itself?

Not by default. Continuum starts in learn mode; enforce mode is a later graduation you configure, per the security blog.

Do I need Kiro?

No. Kiro is one IDE hook. GitHub/GitLab/Bitbucket, Claude Code, and MCP are the other doors listed in the Security Agent post.

Should a clinic turn on enforce mode?

No, not until staging, backups, and a named reviewer exist — the same caution as the state of small-business automation.

Where does this sit in a business workflow?

As a proposer on an existing change ticket, next to pricing for the workflow that already has intake and review, not as a silent merge bot.

What to do next

If you are not on AWS or not in the preview, skip the waitlist. Write down your three riskiest internet-facing apps and who is allowed to merge to them. If you are in the preview, stay in learn mode, dump threat models into the ticket, and merge nothing you did not run on staging.

When that ticket already lives in an agentic workflow, Continuum is another scanner on the same queue. Keep the stop button.