Security basics
How basic security stops AI-accelerated attacks
AI makes attackers faster and their lures more convincing, but it mostly speeds up the same routes in: stolen sign-ins, unpatched software, phishing and over-shared data. Lavawall® monitors the basic controls that close those routes across Microsoft 365, Google Workspace, computers and websites, 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| Automated scanning of websites | Fix exposed and outdated web services | Domain scan, WordPress checks |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 | Agentic AI discovery, SaaS Discovery |
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. The trail lets you answer "who touched this?" in seconds and 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. Their 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. It also turns a quarterly good intention into a routine with proof.
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 with the days above their usual range marked. User activity anomaly detection makes a stolen session or a quiet data grab stand out against that person's own normal, so you review fewer false alarms and catch more of the real ones early.
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. It also alerts on MFA method changes, failed MFA and MFA fatigue attempts. 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, and the Canadian Centre for Cyber Security recommends phishing-resistant MFA. Microsoft Entra authentication strengths can require phishing-resistant methods in Conditional Access.
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, and it handles reports of suspect email. In the same plan, security awareness training and phishing simulation let staff practise on safe lures, including real phishing emails turned into lessons, so the pattern is familiar before the real one arrives. Staff who report a Lavawall simulation hear straight away that they caught it.
Unused accounts
Accounts nobody uses keep access nobody reviews, and a stolen password for one can be tried quietly. Unused account monitoring lists accounts inactive for 30, 90 or 365 days or never used, across Microsoft 365, Active Directory and Google Workspace. Lavawall also alerts when a dormant account becomes active again. Approvers can then remove what is not needed, and the decision is kept as evidence. It is among the cheapest ways to shrink what an attacker can try.
Agentic AI monitoring
An AI agent on a computer can open files and run commands. Agentic AI discovery 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. SaaS Discovery finds the AI services people sign up for. 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, so expect more frequent updates. 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 AI is helping defenders find more issues, so each release will carry more fixes. 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, and the security news ticker keeps advisories from national cyber centres in view, so urgent fixes are not missed. Each device, application and update is tracked with a timestamp, ready to export for an auditor or insurer.
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. Fix the high-risk findings first, then schedule rescans on a paid plan so you hear when something changes. Scan only domains you own or have permission to assess.
A baseline checklist
- Every account has MFA, and 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.
Last reviewed . Sources re-checked against the publishers' pages on that date.
Frequently asked questions
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.
Why does AI make patching more urgent?
AI can shorten the time between a vulnerability becoming known and an attacker using it. The Canadian Centre for Cyber Security says frontier AI models can help attackers find and exploit vulnerabilities, which shortens the time defenders have to respond, and recommends preparing for an increase in patching frequency. Lavawall monitors patch status across Windows, macOS and Linux.
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 lists accounts with no MFA or weak MFA, alerts on MFA changes and MFA fatigue, and watches Conditional Access so risky flows can be blocked.
How does Lavawall help MSPs with AI-accelerated attacks?
Lavawall gives MSPs one multi-tenant console to monitor file activity, access approvals, Conditional Access, MFA, unused accounts, user activity, phishing reports, patching, exposed websites and AI tool use across every client. It reports and alerts; it does not claim to stop every attack.
Does Lavawall protect AI models or applications we build?
No. Lavawall secures how your organization uses AI tools. It does not test or protect AI models or applications you build, and it does not filter prompts or block AI traffic.
What is the quickest way to start?
Run the free Lavawall domain scan to see your website, email and DNS the way an attacker sees them, then start a free trial to connect Microsoft 365 or Google Workspace and review the baseline checklist on this page.
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.