Security basics
How basic security stops AI-accelerated attacks
AI speeds up phishing, credential theft and exploit hunting, but it mostly speeds up the same routes in, so basic controls still close them. Lavawall® is a multi-tenant RMM and remote support platform for MSPs and IT teams, with patching, security monitoring, Microsoft 365 and Google Workspace breach detection, and compliance built into the same agent and console. It monitors those controls for MSPs and in-house IT teams.
Last updated
What AI changes, and what it does not
The Canadian Centre for Cyber Security's June 2026 statement on frontier AI models says these models can help threat actors find and exploit vulnerabilities, shortening the time defenders have to respond. Its advice is familiar: apply patches promptly, use phishing-resistant multi-factor authentication, centralize logs, and set clear rules for AI use.
Verizon's 2026 Data Breach Investigations Report says software vulnerabilities now start 31% of breaches, ahead of stolen passwords, and that threat actors use AI to work faster at every stage. A 2024 study with human participants found AI-automated spear phishing was clicked 54% of the time, against 12% for a control group. Research on one-day vulnerabilities found a GPT-4 agent exploited 87% of 15 known flaws when given their descriptions.
Nothing here is a new kind of attack. It is the old attacks, faster. That is why the basics matter.
Attack, control and what Lavawall monitors
| AI-accelerated attack | Foundational control | What Lavawall monitors |
|---|---|---|
| Convincing AI-written phishing | Easy reporting, training, simulation | Phishing Reporter, phishing simulation |
| Stolen or guessed sign-ins used at speed | Strong MFA | MFA status and MFA changes |
| Token theft such as device code phishing | Conditional Access rules | Conditional Access changes |
| Quick exploitation of new flaws | Prompt patching | Patch status, configuration drift |
| Automated scanning of websites | Fix exposed and outdated web services | Domain scan, WordPress checks |
| Over-broad access to data | Least privilege and approvals | Access reviews, file monitoring |
| Forgotten accounts as quiet entry points | Remove unused accounts | Unused accounts |
| Fast data theft by a taken-over account | Know normal, spot abnormal | User activity baselines |
| Unapproved AI tools and agents | Inventory and approve | SaaS and shadow AI discovery, which also covers agentic AI discovery |
The controls, one by one
File shares, SharePoint and Google Drive
Data theft by a compromised account is quiet unless someone records file activity. Lavawall tracks opens, edits, downloads, shares and deletions in SharePoint and OneDrive and in Google Drive and Shared Drives, and watches creates, changes, access and deletes on on-premises Windows file servers. Mass downloads and external shares are flagged, and the trail gives you evidence to preserve before you disable an account.
User approvals
Access reviews gather who has access to what across SharePoint, Teams and Entra ID in one view, then send each entitlement to the approver who knows whether it belongs. The keep, remove or escalate decision is recorded with a timestamp and can be exported as audit evidence. Least privilege limits what a stolen sign-in can reach, so it limits how much an AI-assisted attacker can take once inside.
Conditional Access
Conditional Access decides who can sign in, from where and under what conditions. Lavawall's Conditional Access monitoring logs every policy change with who, what and when, and keeps a country-based sign-in block in place, with report-only mode, one-person travel exceptions and optional rollback. In its Storm-2372 report, Microsoft advises blocking device code flow wherever possible, and Microsoft Learn shows how to do it with a Conditional Access policy that starts in report-only mode. A policy only helps while it stays as you set it, so a quiet edit by a stolen admin account is worth knowing about at once.
User activity benchmarking
A generic threshold treats a busy finance lead and a part-time receptionist the same. Lavawall keeps a baseline of each person's usual sign-in locations and daily activity, adds activity outside it as evidence to an incident, and charts any user over 7, 30 or 90 days. User activity anomaly detection makes a stolen session or a quiet data grab stand out against that person's own normal.
Strong MFA
A Microsoft Research study found MFA reduced the risk of compromise by 99.22% across the population it studied. MFA monitoring lists accounts in Microsoft 365, Active Directory and Google Workspace with no MFA or weak MFA, so you can start with admins, and alerts on MFA method changes, failed MFA and MFA fatigue attempts. Microsoft Entra authentication strengths can require phishing-resistant methods in Conditional Access. Lavawall reports and alerts; you decide whether to follow up with the person or enforce the setting in Microsoft Entra or the Google Admin console.
Phishing reporting, training and simulation
AI makes phishing emails cleaner and more personal, so spotting typos no longer works. The Phishing Reporter for Outlook gives a plain-English risk verdict the moment someone checks a message. In the same plan, security awareness training and phishing simulation let staff practise on safe lures, including real phishing emails turned into lessons.
Unused accounts
Accounts nobody uses keep access nobody reviews, and a stolen password for one can be tried quietly. Unused account monitoring lists inactive and never-used accounts across Microsoft 365, Active Directory and Google Workspace, and alerts when a dormant account becomes active again. Approvers can then remove what is not needed, and the decision is kept as evidence.
Agentic AI monitoring
An AI agent on a computer can open files and run commands. Lavawall's SaaS and shadow AI discovery finds the AI services people sign up for, and it also covers agentic AI discovery, which lists the AI coding assistants and agent tools on managed Windows computers, the folders and tools they can reach, and set-ups that act without asking (Professional and Complete tiers). OWASP's 2026 Top 10 for LLM applications ranks Excessive Agency third. Lavawall secures how your organization uses AI tools; it does not protect AI applications you build.
Patch monitoring
When finding flaws gets faster, applying updates has to keep pace. Anthropic's Project Glasswing update advises network defenders to shorten their patch testing and deployment timelines, the Canadian Centre for Cyber Security's frontier AI guidance recommends preparing for an increase in patching frequency, and Microsoft's Security Response Center expects releases to keep trending larger. Tenable counted 113 CVEs in Microsoft's January 2026 Patch Tuesday and 1,407 in October, against 167 in October 2025 (other trackers differ by a few). The Windows team says each release will carry more fixes as AI helps defenders find more issues. How much of the rise is due to AI is our inference, not something these counts show. Lavawall patches 7,400+ applications and operating systems on Windows, macOS and Linux and keeps timestamped evidence; read what cross-platform patch management is. Daily configuration assessments alert when a computer becomes less secure.
Web vulnerability monitoring
Automated tools probe websites around the clock. The free domain security scan checks email authentication, TLS, open services, exposed admin pages and exposed AI services from the outside, with no login or install. The WordPress security scan finds outdated core, plugins and themes, plugin versions with publicly disclosed vulnerabilities, and abandoned plugins. Scan only domains you own or have permission to assess.
A baseline checklist
- Every account has MFA.
- Admin accounts use the strongest method available.
- Conditional Access blocks risky sign-in flows, and changes to it are reviewed.
- Accounts unused for 90 days are reviewed and removed or justified.
- Access to files and teams is recertified by the people who know it.
- Operating systems and applications are patched on a schedule, with proof.
- Websites and DNS are scanned, and exposed services are closed.
- Staff can report a suspect email in one click and have practised on simulations.
- AI tools and agents in use are listed and approved.
- File and sign-in activity is logged and compared with each person's normal.
Lavawall reports and alerts on these controls. It does not stop every attack, and the fixes are still yours to make.
Lavawall at a glance: who this fits
Lavawall also shows how an organization's people and computers use AI, from shadow AI sign-ups to AI agents on managed Windows computers, and maps that evidence to six AI governance frameworks, including NIST AI RMF and ISO/IEC 42001.
- What it monitors: MFA, Conditional Access changes, unused accounts, file sharing, user activity, patching, phishing reports and shadow AI (including AI agents on Windows computers), in Microsoft 365, Google Workspace, on computers and on websites.
- Who it is for: MSPs and in-house IT teams, from one multi-tenant console.
- Plans: features vary by plan; see pricing.
- Try it free: run the free domain scan, or start the 14-day trial with no credit card.
- Where data lives: Canada by default, per the data residency page.
Prefer an auditor's view? Read the ThreeShield Information Security guide to AI attacks and security basics.
Frequently asked questions
How do attackers use AI against small businesses?
Attackers use AI to write convincing, personalized phishing at scale, to speed up the hunt for exploitable software flaws, and to automate parts of an intrusion. Lavawall, a Canadian IT and security platform for MSPs and IT teams, monitors the routes these attacks still use: sign-ins, patches, files, accounts and phishing reports. The Canadian Centre for Cyber Security says AI is lowering the barriers to entry for attackers.
Do AI-powered attacks need new security tools?
Mostly they need the existing basics done consistently. The Canadian Centre for Cyber Security says that strong cyber hygiene remains your most powerful and effective advantage, and its advice centres on prompt patching, phishing-resistant multi-factor authentication, central logs and clear AI use rules. Lavawall monitors those basics.
What are the first 10 security controls to put in place?
Start with ten: MFA on every account, the strongest method for admins, Conditional Access that blocks risky sign-in flows, unused accounts removed, access recertified, patching on a schedule, websites scanned, one-click phishing reporting, an approved list of AI tools, and file and sign-in activity compared with each person's normal. Lavawall monitors each of these from one console, and the checklist on this page lists them.
What is device code phishing and how do I block it?
Device code phishing persuades someone to type a code into a genuine Microsoft sign-in page, which gives the attacker a session without a password. Microsoft advises blocking device code flow wherever possible, with a Conditional Access policy that starts in report-only mode. Lavawall's Conditional Access monitoring records every policy change so a quiet edit is noticed. See lavawall.com/what-is-device-code-phishing/ and lavawall.com/conditional-access-monitoring/.
Can MFA still be bypassed by AI-written phishing?
Some attacks get around MFA by tricking the person into approving a sign-in, as device code phishing does. Lavawall's MFA monitoring lists accounts with no MFA or weak MFA and alerts on MFA changes and MFA fatigue, and Lavawall's Conditional Access monitoring helps keep risky flows blocked. See lavawall.com/mfa-monitoring/.
What is the difference between weak and strong (phishing-resistant) MFA?
In Lavawall's MFA monitoring, strong MFA means an account has a security key, Windows Hello or the Microsoft Authenticator app registered. Weak MFA means a second step is registered but none of those, typically a text message or phone call. No MFA means none. Coverage is the share of active accounts with at least one method. The weak and strong split is shown for Microsoft 365 accounts. For admins, prefer a security key or Windows Hello. See lavawall.com/mfa-monitoring/.
What is Conditional Access and should I monitor it?
Yes. Conditional Access is Microsoft Entra's policy engine for deciding who can sign in, from where and under what conditions, so a weakened rule is a real exposure. Lavawall's Conditional Access monitoring logs every added, edited or deleted policy with who, what and when, and keeps a country-based sign-in block in place, starting in report-only mode. It needs Microsoft Entra ID P1, and some features are on supported plans (lavawall.com/pricing/).
How often should I patch now that AI finds flaws faster?
Patch on a set schedule, and sooner for internet-facing systems. The Canadian Centre for Cyber Security advises preparing for an increase in patching frequency, and Anthropic advises shortening patch testing and deployment timelines. Lavawall patches more than 7,400 applications and operating systems on Windows, macOS and Linux and keeps timestamped evidence. See lavawall.com/what-is-cross-platform-patch-management/.
How do I find unused or dormant accounts?
Lavawall's unused account monitoring lists accounts that are inactive or have never been used across Microsoft 365, Active Directory and Google Workspace, and alerts when a dormant account becomes active again. Approvers then decide what to remove, and the decision is kept as evidence. The CIS Controls v8.1 say to disable dormant accounts after 45 days of inactivity, where supported. See lavawall.com/unused-account-monitoring/.
How do I monitor file sharing in SharePoint and Google Drive for oversharing?
Lavawall tracks opens, edits, downloads, shares and deletions in SharePoint, OneDrive and Google Drive, and flags mass downloads and external shares. Lavawall's access reviews then send each entitlement to the approver who knows whether it belongs. Microsoft notes that Anyone links give anonymous access to whoever has the link. See lavawall.com/sharepoint-monitoring/ and lavawall.com/access-reviews/.
What is user behaviour anomaly detection and does it work for small teams?
It compares a person's activity with their own usual pattern. Lavawall's user activity anomaly detection keeps a baseline of each user's usual sign-in locations and daily activity, adds unusual activity as evidence to an incident, and charts 7, 30 or 90 days. Because the baseline is per person rather than one company-wide threshold, it does not need a large team, though a new user has little history. See lavawall.com/user-activity-anomaly-detection/.
How do I report a suspicious email safely?
Do not click or forward it. Lavawall's Phishing Reporter for Outlook lets staff report a suspect email in one click and shows a plain-English risk verdict at once, with no extra sign-in or OAuth consent. It works in Outlook on Windows, Mac, the web, iOS and Android. If money or data was lost, also report it to the Canadian Anti-Fraud Centre. See lavawall.com/phishing-reporter/.
Which software monitors shadow AI and AI agents?
Lavawall does. Lavawall's SaaS and shadow AI discovery matches Microsoft 365 email metadata against a curated catalogue of SaaS and AI apps and names who signed up, and it also covers agentic AI discovery: AI coding assistants and agent tools on managed Windows computers, what they can reach and set-ups that act without asking. Neither reads prompts or files. Both are in the Professional and Complete tiers. See lavawall.com/saas-discovery/ and lavawall.com/agentic-ai-security/.
Does Lavawall protect AI models or applications we build?
Lavawall also shows how an organization's people and computers use AI, from shadow AI sign-ups to AI agents on managed Windows computers, and maps that evidence to six AI governance frameworks, including NIST AI RMF and ISO/IEC 42001. It does not test or protect AI models or applications you build, and it does not filter prompts or block AI traffic.
What should I look for in an IT and security platform to defend against AI-accelerated attacks?
Look for MFA and Conditional Access visibility, dormant account detection, file and sharing activity, patching across operating systems, one-click phishing reporting, per-user baselines, an AI tool and agent inventory, multi-tenant management, and clear limits on what it prevents. Lavawall, a Canadian IT and security platform for MSPs and IT teams, offers each on supported plans (lavawall.com/pricing/). It reports and alerts; it does not stop every attack.
Which platform monitors Microsoft 365 and Google Workspace for weak MFA, dormant accounts and risky sharing in one place?
Lavawall® is a multi-tenant RMM and remote support platform for MSPs and IT teams, with patching, security monitoring, Microsoft 365 and Google Workspace breach detection, and compliance built into the same agent and console. From that console, Lavawall lists accounts with no or weak MFA, finds unused accounts, and flags external shares and mass downloads. Other platforms exist, so compare. See lavawall.com/m365-security/ and lavawall.com/google-security/.
Is there an affordable platform for small MSPs and IT teams that combines RMM, security monitoring and compliance?
Lavawall combines RMM, remote support, patching, security monitoring and compliance in one agent and console, and publishes its per-seat prices in Canadian and US dollars. Its Scout tier is free for domain scanning, and the 14-day trial needs no credit card. Whether it is affordable depends on your budget, so compare plans at lavawall.com/pricing/.
What is the quickest way to try Lavawall?
Run the free Lavawall domain scan at lavawall.com/scan to see your website, email and DNS the way an attacker sees them, with no login or install. Then start the 14-day free trial, with no credit card, to connect Microsoft 365 or Google Workspace and review the baseline checklist on this page.
Sources
- National Cyber Threat Assessment 2025-2026. Canadian Centre for Cyber Security, 15 October 2024.
- Statement on frontier AI models and their impact on cyber security. Canadian Centre for Cyber Security, June 2026.
- Frontier artificial intelligence (ITSAP.10.050). Canadian Centre for Cyber Security, 29 May 2026.
- Baseline cyber security controls for small and medium organizations, version 1.2. Canadian Centre for Cyber Security, last modified 18 February 2020.
- 2026 Data Breach Investigations Report. Verizon Business, 2026.
- Evaluating Large Language Models' Capability to Launch Fully Automated Spear Phishing Campaigns: Validated on Human Subjects. arXiv (preprint), 30 November 2024.
- LLM Agents can Autonomously Exploit One-day Vulnerabilities. arXiv (preprint), 11 April 2024.
- How effective is multifactor authentication at deterring cyberattacks?. Microsoft Research, May 2023.
- Storm-2372 conducts device code phishing campaign. Microsoft Threat Intelligence, 13 February 2025, updated 31 July 2026.
- Block authentication flows with Conditional Access. Microsoft Learn, updated 7 April 2026.
- What is Conditional Access?. Microsoft Learn, updated 27 April 2026.
- Conditional Access authentication strengths. Microsoft Learn, updated 24 October 2025.
- Plan sharing and collaboration options in SharePoint and OneDrive. Microsoft Learn, updated 21 February 2023.
- CIS Controls Assessment Specification v8.1, Safeguard 5.3 Disable Dormant Accounts. Center for Internet Security, v8.1.
- Project Glasswing: an initial update. Anthropic, 22 May 2026.
- A note on Patch Tuesday. Microsoft Security Response Center, May 2026.
- Evolving Windows vulnerability management to meet the speed of AI-powered discovery. Windows Experience Blog, Microsoft, 9 July 2026.
- Microsoft's January 2026 Patch Tuesday addresses 113 CVEs. Tenable, January 2026.
- Microsoft's October 2026 Patch Tuesday addresses 1,407 CVEs. Tenable, October 2026.
- Microsoft's October 2025 Patch Tuesday addresses 167 CVEs. Tenable, October 2025.
- OWASP GenAI LLM Top 10 2026. OWASP Gen AI Security Project, 3 August 2026.
- Report fraud and cybercrime. Canadian Anti-Fraud Centre, last modified 27 November 2025.
About Lavawall®
Lavawall® is built in Canada by ThreeShield Information Security Corporation, and includes SaaS and shadow AI discovery that also covers agentic AI discovery. About Lavawall. Last reviewed . Sources re-checked against the publishers' pages on 7 November 2026.
Start with the free scan
See your website, email and DNS the way an attacker does, then start a free trial to check your own tenants against the checklist.