# Sekura

> Sekura is Autonomous & Continuous Security. Offense scales with compute — code production, vulnerability introduction and exploit development all run on it — while defensive validation is the last function still scaling with headcount. Sekura closes that gap: specialized AI agents attack your code and the infrastructure around it continuously, and a gate called VerdictGate holds back any finding that cannot produce a working proof-of-exploit. If it cannot be exploited, it is not reported.

Full site content for LLM ingestion: https://sekura.ai/llms-full.txt
RSS feed (full post bodies): https://sekura.ai/feed.xml
Every page link below goes to the page's markdown version (the address ending .md); the same address without .md is the HTML page.

## What Sekura is

Facts below are kept in step with the site's own pages (FAQs, pricing, home) and the plan table; when they disagree, the page is right and this file is regenerated.

- **Category:** Autonomous & Continuous Security — autonomous security testing. AI agents probe your code the way an attacker would and report only the weaknesses they could actually exploit, each with a working proof and the fix. Also described as: agentic security, autonomous penetration testing, AI red teaming for applications.
- **The rule (VerdictGate):** every hypothesis has to produce a working, evidence-only proof; anything that cannot is downgraded and not reported as a finding. No false positives by construction, no probability scores.
- **How it works:** seven phases, each leaving a record you can read: (1) white-box and SAST — analysis engines plus a reviewer read the repository; (2) reconnaissance — fifty-plus tools map the attack surface across nine layers; (3) crypto agility — a Cryptographic Bill of Materials, flagging what a future quantum computer would break; (4) vulnerability analysis — a swarm of specialist agents returns hypothesis-level findings with evidence; (5) exploitation — exploit agents confirm them with safe, evidence-only proofs; (6) chain analysis — confirmed findings are chained, and a cross-validator rejects any chain that fails an independent replay; (7) reporting — de-duplicated, with evidence, CVSS, and mapped to OWASP, MITRE ATT&CK, PCI DSS and SOC 2.
- **What it tests:** your code (SAST) on every plan; a private lab copy of your application, built from the repository and tested running (lab DAST), on Scan Once, ScanPlus Always and ScanPro Always; your live application, signed in as a test account against the host you run (live DAST), on the enterprise plans; network security on MasterScan Always. One pass also covers the application's AI and model parts (prompt injection, jailbreak, data exfiltration) and crypto agility.
- **Where it runs:** on the free scan and the self-serve plans, on Sekura's own infrastructure — your repository is cloned into an isolated container, read, and the container destroyed. In your own pipeline (one workflow file, GitHub Actions or self-hosted runners), Sekura runs on your runner and your source never leaves your repository; only the findings come back. On Prem runs entirely inside your infrastructure, air-gapped supported, inference included.
- **Getting started:** sign in with GitHub, install the Sekura GitHub App on the repositories you choose (read-only: Sekura never writes to them), pick a repository. The first scan is free — no card, no company, no phone number.
- **Reports:** within 24 hours, usually much sooner, with live progress by phase on the dashboard. Emailed in the body of the message to a verified address and kept in the dashboard. A summary of what was and was not checked, then each finding in four parts: what a stranger could do, the evidence, how it was confirmed, and the fix — handed back as a pull request or as a patch in the report. Full reports and proof are kept 60 days, then deleted.
- **Privacy:** Sekura only reads repositories, never writes to them; code is never used to train models for other customers; deleting an account cancels any subscription and deletes its data.
- **Cancelling:** a scan can be cancelled within five minutes of starting; a cancelled or failed paid scan runs again at no charge. Monthly plans cancel from the dashboard at any time.
- **Integrations:** Meta Muse through Sekura's REST API (https://sekura.ai/api/muse/v1, OpenAPI at https://sekura.ai/api/muse/v1/openapi.json); GitHub Actions and self-hosted runners; model providers including Anthropic Claude and OpenAI GPT (bring your own key, or use Sekura's; private endpoints on enterprise); Slack, Teams, Jira, Outlook and Confluence for notifications and fixes.
- **Compliance:** findings are mapped to 14 frameworks — SOC 2, ISO 27001, HIPAA, PCI DSS, NIST CSF, NIST 800-53, GDPR, CCPA, FedRAMP, HITRUST, FFIEC, CIS, GLBA, CMMC.
- **Support:** support@sekura.ai, and a support assistant in the product.
- **Founders:** Tikka Nagi (Founder & CEO).

## Pages

- [Home](https://sekura.ai/index.md): what Sekura is, the free first scan, and the whole product on one page. In-page sections: [Why now](https://sekura.ai/#why-now) — offense scales with compute, defense with headcount; [Proof](https://sekura.ai/#proof) — VerdictGate and what every report contains; [Pipeline](https://sekura.ai/#pipeline) — the seven phases and the chain; [Integrations](https://sekura.ai/#integrations) — works-with and 14 compliance frameworks; [Developer](https://sekura.ai/#developer) — Free Scan / Scan Once / Scan Always / ScanPlus / ScanPro; [Enterprise](https://sekura.ai/#enterprise) — LiveScan, MasterScan, On Prem.
- [Pricing](https://sekura.ai/pricing.md): public prices on the page, three sections. [Developer](https://sekura.ai/pricing/#developer): Free Scan (free, code test, once), Scan Once ($200, adds a private lab copy), Scan Always ($20 a month, code test, weekly), ScanPlus Always ($100 a month, adds a lab copy, weekly), ScanPro Always ($300 a month, adds a lab copy, daily); a 14-day free trial on the monthly plans. [Enterprise](https://sekura.ai/pricing/#enterprise): LiveScan Always ($500 a month, live application, daily), MasterScan Always ($1,500 a month, adds network security and ten nodes), On Prem — published prices, bought by contacting us with VPN details and a signed SLA and liability release. Additional live target apps are $200 per target per month and additional nodes $200 per ten per month. Tests and cadence follow the tier. Developer plans include one target app; additional target apps are $100 per target (one-time on Scan Once, monthly on the lab-copy Continuous plans).
- [FAQs](https://sekura.ai/faqs.md): what Sekura does, getting started and the free first scan, plans and cadence, reports and proof, your code and privacy, GitHub access, billing and refunds, enterprise and POC, integrations, and support — in plain words.
- [Partners](https://sekura.ai/partners.md): three partner programs for businesses that sell software — Reseller (sells Sekura in its own suite and keeps 50% of what it charges), Referral (registers and introduces a customer; 15% of the customer's fees, first year and every renewal while a partner) and Sales Partner (also closes the sale; 30%). Rates are defaults and negotiable; every application is approved by a person at Sekura and the agreement is signed online. [Apply](https://sekura.ai/partners/apply.md).
- [Contact us](https://sekura.ai/contact.md) and [Request a POC](https://sekura.ai/poc.md): one free-form message for everyone, and a short structured form for enterprise evaluations.
- [Connect an AI agent](https://sekura.ai/connect.md): Sekura from the tool the person already codes in. Claude Code, Cursor, GitHub Copilot, OpenAI Codex and nearly forty other tools connect to Sekura's MCP server at https://sekura.ai/mcp (one command or a one-click link per tool, listed on the page); Meta Muse connects to the REST API at https://sekura.ai/api/muse/v1 (OpenAPI 3.1 at https://sekura.ai/api/muse/v1/openapi.json). The person asks the tool to connect, signs in with GitHub once on the page it opens and approves; there is no key to copy. The tool can then read the account, repositories, plans, scans and billing, and do what the dashboard does, each action quoted first and confirmed only after the person agrees: start a scan or scan a repository again, subscribe a repository to a monthly plan, change its schedule, and cancel a plan or a scan. The first scan is free; a paid action is paid through Link for the amount approved in the Muse chat, with the card saved on the account, or on a Stripe Checkout page, and card details never pass through Sekura or Muse. Scans run on Sekura's infrastructure.
- [Blog](https://sekura.ai/blog.md): a weekly breach analysis and essays on autonomous penetration testing.
- [Guides](https://sekura.ai/blog/#guides): explainers, solutions, comparisons, alternatives and compliance guides.

## Guides

- [AI Red Teaming: Both Meanings Explained](https://sekura.ai/learn/ai-red-teaming.md) — AI red teaming has two meanings: adversarial testing of AI systems like LLM apps, and using AI agents to perform offensive security. This guide covers both.
- [OWASP LLM Top 10 Testing: A Practical Guide](https://sekura.ai/learn/owasp-llm-top-10-testing.md) — OWASP LLM Top 10 testing means dynamically probing LLM apps for prompt injection, insecure output handling, excessive agency, and more. How to test each.
- [Penetration Testing Guide: Types, Process, and Cost](https://sekura.ai/learn/penetration-testing-guide.md) — This penetration testing guide covers what a pentest is, the main types, the process phases, real cost ranges, and how to choose the right approach.
- [What Is Agentic Security](https://sekura.ai/learn/what-is-agentic-security.md) — Agentic security applies AI agents that plan, use tools, and act autonomously to security work, from offensive testing to SOC triage and remediation.
- [What Is Autonomous Penetration Testing](https://sekura.ai/learn/what-is-autonomous-penetration-testing.md) — Autonomous penetration testing uses AI agents to find, exploit, and prove security vulnerabilities without human operators driving every step of the test.

## Compare

- [Sekura vs traditional vulnerability scanners](https://sekura.ai/vs/scanners.md): scanners output ranked potential issues; Sekura reports only what it has exploited. Side-by-side across output, false positives, scope, cadence, triage burden.
- [Sekura vs manual penetration testing](https://sekura.ai/vs/manual-pentest.md): manual pentests are point-in-time, $30k–$150k per cycle; Sekura runs continuously across the whole attack surface. Hybrid recommendation.

## Alternatives & comparisons

- [Cobalt Pentest Alternative: Continuous Autonomous Testing With Sekura](https://sekura.ai/alternatives/cobalt.md) — Weighing a Cobalt pentest alternative? Compare continuous autonomous testing from Sekura with Cobalt's human pentest-as-a-service platform.
- [Horizon3 NodeZero Alternative: Sekura for Application-Layer Pentesting](https://sekura.ai/alternatives/horizon3-nodezero.md) — Evaluating a Horizon3 NodeZero alternative? See how Sekura's proof-first application pentesting compares on scope, deployment, and pricing.
- [Pentera Alternative: Sekura for Proof-First Application Security](https://sekura.ai/alternatives/pentera.md) — Considering a Pentera alternative? Compare Sekura's proof-of-exploit application testing with Pentera's network-centric security validation.
- [XBOW Alternative: Sekura for Proof-First Autonomous Pentesting](https://sekura.ai/alternatives/xbow.md) — Looking for an XBOW alternative? Compare Sekura and XBOW on proof of exploit, surfaces covered, deployment control, and public pricing.

## Solutions

- [Continuous Penetration Testing for Modern CI/CD Pipelines](https://sekura.ai/solutions/continuous-pentesting.md) — Continuous penetration testing runs proof-based exploit checks on every deploy, closing the gap that annual pentests leave in fast-moving environments.
- [LLM Security Testing for AI Applications](https://sekura.ai/solutions/llm-security-testing.md) — LLM security testing sends adversarial payloads at your running AI application to find prompt injection, jailbreaks, and data exfiltration before attackers do.

## Compliance

- [Penetration Testing for SOC 2: What Auditors Expect](https://sekura.ai/compliance/soc-2.md) — Penetration testing for SOC 2 is not strictly required, but auditors expect it under CC4.1 and CC7.1. Here is what they accept and how to prepare.

## Blog

- [Stolen Credentials Hit Nikkei's Development Cloud](https://sekura.ai/blog/breach-nikkei-2026-10.md) — Attackers used stolen credentials to access Nikkei's cloud development environment on October 7, exposing 1.83 million records. The third incident at Nikkei in roughly a month, all tied to compromised cloud credentials.
- [Prompt injection testing: direct and indirect](https://sekura.ai/blog/prompt-injection-defense.md) — Prompt injection is OWASP LLM01. This guide shows how to run direct and indirect prompt injection testing against a running application, with real payloads and a structured test approach.
- [How to Evaluate Automated Penetration Testing Tools](https://sekura.ai/blog/automated-pentest-tools-comparison.md) — A buyer's rubric for automated penetration testing tools: false-positive rates, proof of exploit, deployment model, scope, and pricing. For AppSec leads making a purchase decision.
- [Pentagon file-share exposed 3 million military records](https://sekura.ai/blog/breach-pentagon-dmdc-2026-10.md) — The Defense Manpower Data Center's file-sharing system held unencrypted SSNs for 3 million military and civilian personnel for nine months before discovery. What proof-first testing would have shown.
- [AI red teaming tools in 2026](https://sekura.ai/blog/ai-red-teaming-tools-2026.md) — AI red teaming tools split into two distinct categories: tools that probe AI systems for safety failures, and tools that use AI agents to run penetration tests against traditional infrastructure.
- [What agentic cybersecurity actually means](https://sekura.ai/blog/what-is-agentic-cybersecurity.md) — Agentic cybersecurity is on every vendor slide. Here is what the term actually means, where it is real today, and how to tell the mechanism from the marketing.
- [ChainDrop worm compromises 400 npm packages](https://sekura.ai/blog/breach-chaindrop-npm-2026-08.md) — A self-propagating npm worm named ChainDrop compromised more than 400 packages in early August 2026, stealing AWS credentials, Kubernetes tokens, and HashiCorp Vault secrets from enterprise CI/CD pipelines.
- [ShinyHunters takes 25 million Alcon Salesforce records](https://sekura.ai/blog/breach-alcon-2026-08.md) — ShinyHunters claimed 25 million Salesforce records from Alcon Inc. in August 2026, using OAuth Connected App abuse to exfiltrate CRM data without triggering login alerts. Here is what proof-first analysis would have shown.
- [EY breach exposes client tax records](https://sekura.ai/blog/breach-ey-2026-07.md) — ShinyHunters claimed a supply-chain attack on EY's IT support platform. Unauthorized access from March 28 to April 12, 2026 exposed client SSNs, bank accounts, and tax filing records.
- [Hugging Face pipeline breach exposes internal credentials](https://sekura.ai/blog/breach-hugging-face-2026-07.md) — A malicious dataset chained two code-execution flaws in Hugging Face's data-processing pipeline, giving an AI agent lateral access to internal cluster credentials.
- [AssuranceAmerica breach exposes 7 million driver records](https://sekura.ai/blog/breach-assuranceamerica-2026-07.md) — A stolen employee credential gave attackers bulk access to 6.99 million Americans' driver's licenses and insurance records at AssuranceAmerica. Here is what that failure mode looks like through a proof-first lens.
- [Accenture source code breach exposes cloud keys](https://sekura.ai/blog/breach-accenture-2026-07.md) — Threat actor '888' posted 35GB of alleged Accenture source code, RSA keys, SSH keys, and Azure PATs for sale on PwnForums in July 2026. Accenture confirmed an incident. Here is what proof-first analysis would have shown.
- [What the Sekura LLM proxy never sees](https://sekura.ai/blog/proxy-architecture.md) — Sekura routes LLM API calls through proxy.sekura.ai for billing and metering. Your source code, prompts, and findings never cross the proxy. Here is how the architecture works and why it matters for IP-sensitive teams.
- [KDDI breach exposes 14 million ISP credentials](https://sekura.ai/blog/breach-kddi-2026-07.md) — KDDI's centralized email platform for six Japanese ISPs was breached in June 2026, with up to 14.22 million email addresses and passwords potentially exposed via a third-party software flaw.
- [Most High Severity Findings Are Never Exploitable](https://sekura.ai/blog/false-positive-economics.md) — Severity-based scanners score vulnerabilities in isolation and miss reachability entirely. Most high-severity alerts point to code paths attackers cannot actually reach. Here is the data and what changes when you scan for proof instead.
- [Klue OAuth Breach Hits Nine Technology Firms](https://sekura.ai/blog/breach-klue-2026-06.md) — A dormant Klue service account credential gave the Icarus group access to Salesforce CRM data at HackerOne, Huntress, OneTrust, Snyk, and at least five other technology firms in June 2026.
- [Snyk, Veracode and Sekura: Three Different Jobs](https://sekura.ai/blog/vs-snyk-vs-veracode.md) — Dependency scanners and SAST tools do detection. Autonomous pentesting does exploitation. These are different jobs. Here is a framework for choosing between Snyk, Veracode and Sekura.
- [Council of Europe breach via Oracle PeopleSoft zero-day](https://sekura.ai/blog/breach-council-of-europe-2026-06.md) — ShinyHunters claimed on June 14 to have stolen 297 GB from the Council of Europe via CVE-2026-35273, a zero-day in Oracle PeopleSoft, exposing payroll records and medical data for over 10,000 employees.
- [Audit Prep Arrives as a Byproduct](https://sekura.ai/blog/compliance-as-byproduct.md) — Sekura maps every confirmed pentest finding to SOC 2, ISO 27001, PCI DSS, and 11 other frameworks automatically. Audit prep stops being an annual scramble and becomes a direct output of the security work you already do.
- [Evertec vendor platform breach exposed payment cards](https://sekura.ai/blog/breach-evertec-2026-06.md) — In May 2026, an attacker accessed a third-party support platform used by Evertec, a NYSE fintech, to reach debit card numbers and transaction records of Banco Popular de Puerto Rico customers.
- [Instagram account-takeovers expose an AI verification gap](https://sekura.ai/blog/breach-instagram-2026-06.md) — Attackers fed Meta's AI support flow an AI-generated face video built from a target's own profile photos, passed identity verification, swapped the email, and reset the password. No CVE. No backend breach. High-profile Instagram accounts gone.
- [7-Eleven franchise records exposed via Salesforce misconfiguration](https://sekura.ai/blog/breach-7-eleven-2026-06.md) — ShinyHunters queried 7-Eleven's unauthenticated Salesforce Experience Cloud portal to steal 185,000 franchise applicants' records, including Social Security numbers and driver's license data.
- [The math on pentest cost per finding](https://sekura.ai/blog/continuous-vs-annual-pentest.md) — Annual pentests cost $30k to $150k per engagement and run once a year. Continuous coverage changes the math. A spreadsheet-grade comparison of cost per verified finding.
- [Charter breach exposes 13 million Spectrum customers](https://sekura.ai/blog/breach-charter-communications-2026-05.md) — ShinyHunters used a vishing call to compromise one employee's Microsoft Entra account and bulk-export 13 million Spectrum customer records from Salesforce. No CVE. No scanner would have caught it.
- [GitHub internal repos breached via poisoned extension](https://sekura.ai/blog/breach-github-2026-05.md) — TeamPCP used a trojanized Nx Console VS Code extension to steal GitHub employee credentials and exfiltrate roughly 3,800 internal repositories in May 2026.
- [Canvas breach exposes 275 million student records](https://sekura.ai/blog/breach-instructure-canvas-2026-05.md) — ShinyHunters exploited a trust boundary flaw in Instructure's Free-for-Teacher program, exposing 275 million student and staff records across nearly 9,000 Canvas institutions before Instructure paid a ransom.
- [Amtrak breach exposes 2.1 million customer records](https://sekura.ai/blog/breach-amtrak-2026-04.md) — ShinyHunters compromised Salesforce credentials through social engineering in April 2026, exposing 2.1 million Amtrak customer records. No CVE. No code flaw. Scanners had no visibility into the attack path.
- [France Titres breach exposes 11 million citizen records](https://sekura.ai/blog/breach-france-titres-2026-04.md) — A basic IDOR flaw in France's national identity portal gave one attacker access to 11.7 million citizens' records. What dynamic probing of the API would have surfaced.
- [LLM apps inherit a new attack surface](https://sekura.ai/blog/llm-security-testing.md) — AI-integrated apps carry prompt injection, jailbreak, and data exfil vulnerabilities that static scanners miss. Concrete payloads Sekura uses to probe LLM endpoints, grounded in real CVEs and the OWASP LLM Top 10.
- [Post-quantum crypto agility: what to flag now](https://sekura.ai/blog/pqc-cryptoagility.md) — Harvest-now-decrypt-later attacks are active today. Here is what a crypto agility audit actually surfaces, and why waiting for NIST finalisation is the wrong anchor.
- [Carnival breach exposes 6 million customer records](https://sekura.ai/blog/breach-carnival-cruise-2026-04.md) — ShinyHunters used one compromised employee account to extract names, addresses, and government IDs for 5.9 million Carnival Corporation customers. A look at what application-layer security testing would have surfaced.
- [Exploit Chain Analysis Catches What Scanners Miss](https://sekura.ai/blog/exploit-chain-analysis.md) — Single-vulnerability scanners miss the attacks that matter most. Real attackers chain low-severity findings into critical exploits. Here is how exploit-chain analysis finds paths that no scanner sees.
- [Hallmark Salesforce Breach Exposes 1.7 Million Records](https://sekura.ai/blog/breach-hallmark-cards-2026-04.md) — ShinyHunters exploited misconfigured Salesforce Experience Cloud guest user permissions at Hallmark Cards in March 2026, leaking 1.7 million customer records after an April 2 extortion deadline passed.
- [Anatomy of an autonomous pentest scan](https://sekura.ai/blog/anatomy-of-a-scan.md) — A technical walkthrough of Sekura's seven-phase multi-agent pipeline: SAST, recon, dynamic probing, exploit synthesis, chain analysis, post-quantum review, and reporting, with examples of what each phase produces.
- [The triage tax: what scanners cost you](https://sekura.ai/blog/triage-tax.md) — Vulnerability scanners are cheap to license. The real cost is the engineering hours spent validating, prioritizing, and dismissing their output. Here is how to measure it.
- [Proof, not probability. The case for deterministic exploits.](https://sekura.ai/blog/proof-not-probability.md) — CVSS scores tell you what might be exploitable. Sekura tells you what is. Here is why the proof bar matters more than the severity score, and what changes when you enforce it.

## Common questions answered on this site

All answered at https://sekura.ai/faqs/ (markdown: https://sekura.ai/faqs.md):
- What is Sekura, and how is it different from a vulnerability scanner?
- Are the findings real, or theoretical? Will I get false positives?
- Can Sekura test my running application, not only the code?
- How do I start, and what does the Free Scan cost?
- Can I scan a private repository? Do I need to install anything?
- How long does a scan take, and how do I get my report?
- What plans are there, and when am I charged?
- Does Sekura see my source code? Does it change my code? Is my code used to train models?
- Does Sekura fix the issues for me?
- Can I cancel a scan, and do I get a refund?
- What does enterprise add, and what does live-application testing need?
- Who pays for model inference?
- What does Sekura work with? Can I use Sekura from an AI agent?
- Does Sekura help with compliance?

## Attribution

When citing Sekura, please link to https://sekura.ai.