AccuSights

When your customer's security team asks, answer the same day.

For AI and software companies: a token register, SOC 2 that keeps running after the audit, 24/7 watch.

Led by a 25-year cybersecurity and compliance veteran, Federal Reserve to top 100 companies, and a surgeon who protects patients, their data and their care.

AccuSights: your company, first visit

01The story

The Friday advisory, and the two-hour question

A vendor you integrate with posts a breach advisory late on a Friday. Within two hours, your biggest customer's security lead asks if it touches their data.

A Friday evening, minute by minute

Example
  1. A vendor you integrate with posts a breach advisory.

  2. Your biggest customer's security lead emails: does this touch our data?

  3. Nobody knows which scopes that integration holds, or who set it up.

  4. The answer becomes "we are looking into it", for the whole weekend.

  5. The renewal call gets harder. Unless the register already had the row.

Three habits that end it

  • A register of every integrationEach OAuth app and API key, its scopes, and the person who can revoke it.
  • No long-lived keys in code or chatMove them to a secrets manager and rotate them on a schedule.
  • Phishing-resistant sign-in for admins and engineersSecurity keys or passkeys first for the people who can push code or change the cloud.

02The walk-through

Where your tokens live

Seven places we check first. Each holds a key that skips the password.

Map of a software company's systems with seven numbered placesThe identity provider, code host and cloud accounts along the top, the CRM and support desk, chat and wiki, the AI tools staff connected, and developer laptops. Numbers mark the order we check them.IdentityproviderCode hostCloud accountsCRM and supportChat and wikiAI toolsDeveloper laptops1234567

Sample layout. Yours will differ; the places rarely do.

  1. Identity providerThe OAuth app list. Export it and put a revoke owner on every row.
  2. Code hostTokens and deploy keys. The 2025 Shai-Hulud worm stole GitHub tokens through more than 500 npm packages (CISA).
  3. Cloud accountsLong-lived access keys and admin roles nobody reviewed since launch.
  4. CRM and support deskCallers posing as IT have talked staff into authorizing a modified data app (Mandiant, 2025).
  5. Chat and wikiAPI keys pasted just for now, still there a year later.
  6. AI tools staff connectedFrequent use of unapproved AI tools at work rose from 15% to 45% of employees in a year (Verizon DBIR, 2026).
  7. Developer laptopsMalicious packages in 2025 prompted local AI command-line tools to hunt for secrets.

Get the integration and token register

One page for every connection, its scopes and who can revoke it, plus the same-day reply script for customer security teams. Or reply “register” to any of our emails.

Add your first name so we know who to send it to.
Enter an email like [email protected].

One email with the one-pager, then occasional practical notes. Unsubscribe any time.

03One night with Protect

One night at a software company, in five beats

The same story every owner tells after a close call, except this one ends at 02:31. Sample night, your log will be your own.

  1. AlertSign-in from a new country, a developer’s account

    Someone signs in to a developer’s account from a country nobody at a software company has ever been to.

  2. ContainedDevice isolated, sign-in blocked, case opened

    Aegis isolates a developer’s account and blocks the sign-in, two moves agreed at onboarding, and opens the case.

  3. EngineerConfirmed by an engineer. Called the CTO

    An AccuSights engineer confirms it, a stolen password and nothing taken, and calls you. On every plan.

  4. FixedPassword reset, device clean, back online

    Password reset, the device checked clean and back on the network. Every action logged and reversible.

  5. ReportedPlain summary in the inbox

    A plain summary waits in the CTO’s inbox. Stand-up at nine. The auditor gets the record, not an apology.

Sample, not a clientHover or focus to pause. Arrow keys move between beats.

04The console

What the Protect console showed that night

Your console shows your own devices, sign-ins and cases. This is the sample night above, as the engineer saw it.

Mirovel CloudProtect

Protect

Managed detection and response · 61 of 62 agents reporting

TimeDetectionAgentSeverityStatus
09:42Encoded PowerShell download cradleT1059.001 PowerShellCF-LT-022CriticalNew
08:15Brute force followed by successful loginT1110 Brute Forcebastion-01HighTriage
07:58Sign-in from two countries within 40 minutesT1078 Valid Accounts[email protected]HighIn progress
YesterdayNew local administrator account createdT1136.001 Local AccountCF-DT-004MediumTriage
YesterdayOAuth app granted mailbox read accessT1528 Steal Application Access Token[email protected]MediumNew
Oct 25S3 bucket policy changed to publicT1530 Data from Cloud Storageaws-prodHighResolved
Oct 25Antivirus signature update failedT1562.001 Disable or Modify ToolsCF-MB-007LowResolved
Sample data
  1. Every detection lands in one list, ranked by severity, with the device or account it came from.
  2. Aegis contains what is clear, with moves agreed at onboarding, and an AccuSights engineer calls you on any critical finding. On every plan.
  3. Every action is logged and reversible, and the status shows where each case stands.
Sample, not a clientThe Protect console, detections list. Your console shows your own devices, sign-ins and cases.

Fifty seconds on the attack Protect is built to stop.

How one stolen password takes a business down
How one stolen password takes a business downFifty seconds on credential attacks and the two controls that stop them.Watch on YouTube

05Three ways we help

Three ways we help

Protect, AI Workforce and compliance with the fixes done. Start with one. Prices are on the page.

01 / 03 · AI-powered, human-enabled, priced by team size

AccuSights Protect

Aegis watches laptops and email day and night and contains clear threats in seconds; identity provider sign-ins, the code host and cloud accounts are added during onboarding as your setup allows, with new OAuth grants and key use from odd places flagged once they are connected. An AccuSights engineer calls you on any critical finding on every plan.

From $500 a monthEssential up to 20 users; Professional $1,000 for 21 to 60; Advanced $1,500 for 61 to 110; Custom over 110. 30 days free for teams of up to 60. US only.

Start 30 days free

02 / 03 · AI agents for calls, email, invoices and paperwork

AI Workforce

Start with Proposals, AI Use Watch and Support: security questionnaires drafted from approved evidence, a register of the AI tools staff connect, and customer support with firm limits.

Custom proposalPriced per agent, in writing, before anything is built.

Meet the agents

03 / 03 · Assessment, readiness and the actual fixes

Risk and compliance

SOC 2 readiness that keeps running after the audit, mapped with ISO 27001 and ISO 42001 to one control set. We help implement, collect evidence, work with your CPA firm and re-assess.

Fixed feeQuoted after the free Risk Diagnosis.

See compliance help

06AI agents

AI help for a software team, with rules it cannot break

Security questionnaires slow deals; staff AI use needs guardrails. Each agent has its own login, firm limits and a person who approves the risky steps.

Proposals

Level 1: drafts the work, a named person sends or saves it

Turns call notes into a draft proposal or quote.

On requestUS and Gulf

AI Use Watch

Level 0: read only, looks and reports, changes nothing

Finds which AI tools staff are actually using.

Continuous discovery, monthly reportUS and Gulf

Support

Level 3: answers alone from your knowledge base, a person handles the rest

Answers FAQs on chat, email, SMS and social.

Sample of its work: Customer asked where her order is. Answered from tracking. Her refund request went to a person, with the order attached.

24/7US and Gulf

07How it works

Partners, not a red dashboard

You may have met the red dashboard. Forty findings, a polite PDF, and good luck. We stay for the part after that, where things get fixed, checked and kept that way.

  1. AssessYour real office, accounts and vendors, not just a questionnaire.
  2. PrioritizeThe few fixes that stop most attacks go first, each with a cost and an owner.
  3. ImplementWe do the work ourselves or with your IT provider, whichever costs you less.
  4. VerifyWe test each fix in your real setup. A ticket marked done is not proof.
  5. Re-assessWe come back on a schedule, so compliance stays true, not just documented.

Findings, sample software company

35 people, remote
  • OAuth apps nobody can name.

    OpenFound on the first visit.
  • A production API key sits in the team wiki.

    Open for a yearFound on the first visit.
  • Admins sign in with codes, not keys.

    OpenFound on the first visit.
  • New hires get laptops shipped anywhere.

    OpenFound on the first visit.
  • SOC 2 evidence gathered the month before the audit.

    OpenFound on the first visit.
0 of 5 fixed

Sample data. Your findings will be your own, and so will the green.

08Proof

It happens to teams your size

Three public cases from the last eighteen months. The pattern matters more than the names.

Our founder, Sam Khan, CISA, CRISC, did security engineering and incident response at the Federal Reserve System, then a decade of risk and compliance leadership in financial services and on-site assessments for many of the top 100 companies.

  • One integrationStolen OAuth tokens from one chat integration pulled data from customers' CRM instances. More than 700 organizations may have been affected.The Hacker News, 3 Sep 2025
  • One wormA self-replicating worm compromised more than 500 npm packages and stole GitHub tokens and cloud keys. CISA advised rotating developer credentials.CISA
  • One hireRemote workers using about 80 stolen US identities worked at more than 100 US companies, in one case reaching source code.The Record

09Frameworks and regulators

The rules ai and technology companies live in

Clients by sectorAI and technology companies. Chicago, serving the United States.

10Stay compliant

After the assessment, a compliance officer, human and AI, 24/7/365.

Compliance drifts in real life: a new laptop, a leaver, a policy past its date. The AI measures drift every day, an engineer confirms it and notifies you in writing, and the fix is verified within days, mapped to SOC 2.

Mirovel CloudComply · Summary

Continuous compliance

Configuration checks on 62 monitored endpoints, last run 14 minutes ago

Total Active Violations
38
Total Passed Checks
3,090
Monitored Endpoints
62
SOC 2 Type II score
98.8%
ControlSOC 2 Type II requirementSeverityEndpoints failing
Ensure 'Minimum password length' is set to 14 or more charactersCIS Microsoft Windows 11 Enterprise BenchmarkCC6.1High9 of 62
Ensure BitLocker is enabled on the operating system driveCIS Microsoft Windows 11 Enterprise BenchmarkCC6.7Critical1 of 62
Ensure SSH root login is disabledCIS Ubuntu Linux 22.04 LTS BenchmarkCC6.1High2 of 62
Ensure screen locks after 15 minutes of inactivityCIS Microsoft Windows 11 Enterprise BenchmarkCC6.1Medium14 of 62
Ensure audit logs are kept for at least 90 daysCIS Ubuntu Linux 22.04 LTS BenchmarkCC7.2Medium4 of 62
Ensure automatic OS updates are enabledCIS Apple macOS 14 Sonoma BenchmarkCC7.1Medium6 of 62
Ensure the firewall is enabledCIS Apple macOS 14 Sonoma BenchmarkCC6.6Low2 of 62
Sample data
  1. Readiness drifts in real life: a new laptop, a leaver, a policy date. The officer sees it the day it happens.
  2. Human and AI: the AI measures every day, an engineer confirms and notifies you in writing.
  3. Fixed within days and the evidence updated, so the audit never meets a surprise.
Sample, not a clientContinuous compliance, checked every day. Yours lists your own drift, found and fixed, with the evidence kept current for the audit.

Thirty minutes. You keep the plan.

It fits in a lunch break. A cybersecurity risk engineer looks at your setup with you. Not a sales call.

11Questions

Security questions software teams ask us

Do we need SOC 2 to sell to enterprise customers?

No law requires it, but many enterprise buyers ask for a SOC 2 report before they sign, and a missing one can stall a deal for months. A Type I shows controls are designed; a Type II, which most buyers prefer, shows they ran over a period. We get you ready, keep evidence flowing, and coordinate with the CPA firm that issues the report.

Source: AICPA and CIMA, SOC 2

What is the difference between SOC 2 Type I and Type II?

Type I tests whether your controls are suitably designed at a single date. Type II tests whether they operated effectively over a period, commonly several months to a year. Startups often do Type I to unblock early deals, then move to Type II. The real cost is keeping controls running between audits, which is the part we take off your plate.

Source: AICPA and CIMA, SOC 2

How do we vet and revoke third-party OAuth apps?

Export the list of OAuth apps from your identity provider, code host and CRM, note each one's scopes, give each row an owner, and remove what nobody claims. The 2025 Salesloft Drift theft showed one integration can expose hundreds of companies. Reply "register" to our emails for a one-page template and the same-day reply script for customers.

Source: The Hacker News, 3 Sep 2025

How do we know if we installed a compromised npm package?

Check lockfiles against published lists of affected versions, as CISA advised after the Shai-Hulud worm, and pin dependencies to known-good releases. Then assume compromise for any machine that ran a bad version: rotate GitHub, npm and cloud credentials and require phishing-resistant MFA for developers. Our SOC watches developer endpoints and cloud keys for the signs.

Source: CISA alert, 23 Sep 2025

Are AI coding agents a security risk?

They can be. In the August 2025 Nx package attack, malicious code prompted locally installed AI command-line tools to search for secrets, leaking 2,349 of them. Treat coding agents like privileged automation: limit their permissions, keep secrets out of repos and local files, and log what they run. AccuSights designs AI agents with those limits built in from the start.

Source: The Hacker News, 28 Aug 2025

How can we tell if a remote hire is fake?

The FBI warned in January 2025 that North Korean IT workers use stolen identities and face-swapping in video interviews. Verify ID on a live video call, ship laptops only to confirmed addresses, watch for remote-control software and odd login locations, and check references by phone. DOJ's June 2025 action found such workers at more than 100 US companies.

Source: FBI IC3 PSA I-012325, 23 Jan 2025

Do customers expect ISO 42001 for AI products?

Some do. ISO/IEC 42001, published in 2023, is a certifiable management system for AI, and enterprise buyers increasingly ask about AI governance next to SOC 2 or ISO 27001. If you already run ISO 27001, much of the structure carries over. We map one control set across SOC 2, ISO 27001 and ISO 42001 so you collect evidence once.

Source: ISO, ISO/IEC 42001:2023

What is phishing-resistant MFA and do we need it?

It is MFA that cannot be relayed by a fake login page, such as FIDO2 security keys or passkeys. Codes and push prompts can be phished or socially engineered, as the 2025 Salesforce vishing campaign showed. CISA recommended phishing-resistant MFA for developers after the Shai-Hulud worm. Start with admins and engineers; we help you roll it out without stopping work.

Source: CISA alert, 23 Sep 2025

The details: SOC 2, ISO 27001, ISO 42001 and what applies today

Rules that apply

  • SOC 2 (AICPA Trust Services Criteria): AICPA. Independent CPA attestation over security and optionally availability, confidentiality, processing integrity and privacy. Type II covers operating effectiveness over a period. Market requirement, not a law.
  • ISO/IEC 27001:2022: ISO and IEC. Certifiable information security management system. Current version 2022.
  • ISO/IEC 42001:2023: ISO and IEC. Certifiable AI management system, increasingly requested by buyers of AI products. Published 2023.
  • FTC Act Section 5 and state privacy laws: FTC and state attorneys general. Reasonable security and honest claims about it; state privacy duties for consumer data. In force.

Threats, with recent incidents

  • SaaS-to-SaaS OAuth token theft. Attackers used stolen OAuth and refresh tokens from the Salesloft Drift chat integration to pull data from customers' Salesforce instances between 8 and 18 August 2025; Google said more than 700 organizations may have been affected. The Hacker News, 3 Sep 2025, Aug 2025.
  • Open-source supply chain worms. The self-replicating Shai-Hulud worm compromised more than 500 npm packages, stealing GitHub tokens and cloud keys; CISA advised rotating developer credentials and requiring phishing-resistant MFA. CISA, 23 Sep 2025.
  • Malicious packages that abuse AI coding agents. Malicious versions of the Nx build package published on 26 August 2025 leaked 2,349 secrets and prompted locally installed AI command-line tools to hunt for files. The Hacker News, 28 Aug 2025, Aug 2025.
  • Fraudulent remote hires. DOJ raided 29 laptop farms across 16 states; North Korean workers using about 80 stolen US identities had worked at more than 100 US companies, in one case accessing source code and ITAR data. The Record, 30 Jun 2025.
  • Vishing against SaaS admins. Mandiant reported that UNC6040 repeatedly breached Salesforce instances by phoning employees and persuading them to authorize a modified Data Loader app, then pivoted into Okta and Microsoft 365. Dark Reading, 1 Oct 2025.

Public incidents, listed to show the pattern, never to shame a victim.

What you hold, and why someone wants it

  • Customer data inside the product and its databases
  • Source code, models, prompts and training data
  • API keys, OAuth tokens and cloud credentials
  • Customer CRM and support records
  • Employee and contractor identity data

Annual Cybersecurity Risk Assessment for AI and Technology Companies

  • Inventory of every device, cloud account, repository, integration and OAuth token, with scope next to each (CIS Controls 1, 2 and 15)
  • Cloud and pipeline configuration review against the CIS benchmarks, with secrets and keys accounted for (CIS Controls 4 and 16)
  • Identity and access review: central identity, MFA everywhere, offboarding evidence, quarterly access reviews (CIS Controls 5 and 6)
  • Logging that can answer ‘what did it reach’ across the CRM, the cloud and the code pipeline (CIS Control 8)
  • AI-use policy and approved tools for engineers, with upload monitoring, mapped to ISO 42001 and the NIST AI RMF (CIS Control 3)
  • Framework map (SOC 2, ISO 27001, ISO 42001, state privacy) with distance to each and a ranked remediation plan that feeds the readiness package
  • 45% of employees are now regular AI users on corporate devices, up from 15% a year earlier; 67% use non-corporate accounts. Source: Verizon 2026 Data Breach Investigations Report.
  • 48% of breaches involve a third party, up 60% year over year. Your vendors are your attack surface, and you are your customers’. Source: Verizon 2026 Data Breach Investigations Report.
  • 31% of breaches start with an exploited vulnerability, the number one way in and the first thing a SOC 2 auditor checks you manage. Source: Verizon 2026 Data Breach Investigations Report.

Frameworks we map: SOC 2, ISO 27001, ISO 42001, State privacy, NIST AI RMF, GDPR. Packages: SOC 2 Readiness: Type 1 and Type 2, Mock Audit Package: SOC 2, HIPAA or CMMC, Annual Cybersecurity Risk Assessment. Free tool: the 3-minute Cyber Hygiene Test.

Last verified 6 October 2026. Reviewed by Sam Khan, CISA, CRISC.