Artificial intelligence is changing who can become a serious cyber threat.
Until recently, carrying out a successful business network intrusion required substantial technical knowledge. An attacker needed to understand network reconnaissance, software vulnerabilities, exploit development, credential theft, remote access, data extraction, and methods for hiding their activity.
AI agents are beginning to perform many of those tasks for them.
An inexperienced attacker can now give an AI tool a vague objective and receive technical guidance, computer commands, vulnerability research, custom code, troubleshooting assistance, and recommendations for what to do next. The attacker may still need to make decisions and provide access to tools, but the AI can supply much of the technical structure they previously lacked.
This does not mean every person with an AI account can instantly compromise a business. It does mean that the minimum level of knowledge required to cause significant damage is falling.
For small and midsize businesses in Michigan, that creates an important shift in cybersecurity risk. Organizations are no longer defending themselves only against skilled criminal groups. They must also prepare for opportunistic, inexperienced, and poorly disciplined attackers who can use AI to operate beyond their natural abilities.
A Real-World Case Shows How Quickly the Threat Is Changing
The records reportedly showed an attacker using AI coding agents to assist with reconnaissance, vulnerability research, exploit development, access validation, credential collection, data exfiltration, and reporting. Evidence from the recovered files documented compromises involving at least 14 companies.
What made the case especially concerning was the apparent skill level of the attacker.
The attacker frequently issued vague instructions and depended on the AI agent to determine the technical steps. Instead of personally researching every exposed service, reviewing vulnerability information, building tools, and troubleshooting failed commands, the attacker allowed the AI to perform much of that work.
Researchers concluded that this was not a fully autonomous cyberattack. The human remained involved. However, it was also far more than simple code completion or general advice.
The AI agent functioned as a technical force multiplier.
It helped an inexperienced operator complete tasks normally associated with a much more capable cybercriminal. In some instances, the attacker framed the activity as authorized penetration testing or security research, allowing the malicious requests to resemble legitimate cybersecurity work.
The incident also involved model versions that were not the newest available frontier systems. That distinction matters because AI cyber capabilities continue to develop.
The lesson for business owners is clear: AI-assisted cybercrime is no longer a theoretical future problem. It is already affecting real organizations.
What Are AI-Powered Cyberattacks?
AI-powered cyberattacks are attacks in which criminals use artificial intelligence to improve, automate, accelerate, or scale one or more parts of the attack process.
- Identify businesses with exposed internet-facing systems
- Research known vulnerabilities
- Generate or modify exploit code
- Troubleshoot commands that fail
- Create convincing phishing emails
- Personalize social-engineering messages
- Analyze stolen files
- Search for passwords, financial records, or sensitive information
- Automate repetitive attack tasks
- Translate messages into natural-sounding English
- Produce documentation about compromised systems
- Recommend additional targets or attack paths
AI does not need to invent a completely new form of cyberattack to create a serious problem. Making existing attacks faster, cheaper, more convincing, and easier to execute is enough to increase risk substantially.
AI Is Lowering the Barrier to Cybercrime
The greatest near-term concern is not necessarily that AI will create a completely autonomous super-hacker.
The more immediate problem is that it can help an average attacker perform above their experience level.
Consider the difference between a traditional attacker and an AI-assisted attacker.
A traditional novice might find an exposed server but lack the knowledge to identify its software, research the correct vulnerability, build an exploit, interpret the results, and establish access.
An AI-assisted novice can paste information about the server into an agent and ask it to investigate. The AI may identify the product, research publicly known vulnerabilities, write commands, explain the results, and recommend the next step.
Even when the first attempt fails, the AI can analyze the error and revise its approach.
This creates several consequences for businesses.
1. More People Can Attempt Meaningful Attacks
The population of potential attackers is expanding.
Someone no longer needs years of security training to begin scanning systems, testing stolen credentials, or experimenting with publicly disclosed vulnerabilities. AI can provide on-demand technical assistance throughout the process.
Many of these individuals will still fail. However, the number of attempts matters.
A business that previously faced occasional probing may encounter more automated scanning, credential attacks, phishing attempts, and exploit activity. When thousands of inexperienced att
2. Small Businesses Become More Attractive Targets
Inexperienced attackers often pursue the easiest available opportunity.
They may not begin with a specific interest in a Warren law firm, a Sterling Heights manufacturer, a Macomb County accounting practice, or a Southeast Michigan construction company. They may search the internet for any business exposing a particular firewall, remote-access service, web application, storage device, or outdated server.
The target is often selected because it appears vulnerable.
- Limited internal IT staff
- Outdated software or network equipment
- Weak remote-access controls
- Inconsistent multifactor authentication
- Poor password practices
- Inadequate security monitoring
- Unsupported computers or servers
- Flat networks with limited segmentation
- Backups that have never been tested
- No documented incident-response plan
Attackers do not need to know that a business is wealthy. They only need to see an accessible path into its systems.
3. Known Vulnerabilities May Be Exploited Faster
Software vendors regularly publish updates that correct security vulnerabilities. Once the details become public, defenders and attackers begin working from the same information.
Businesses must install the update before criminals exploit the weakness.
AI can shorten the attacker’s research and development process. It can summarize vulnerability disclosures, analyze proof-of-concept code, adapt scripts, identify affected products, and help troubleshoot exploitation attempts.
This may reduce the time between a vulnerability becoming public and widespread attack activity.
For businesses, patching can no longer be treated as routine maintenance that happens whenever someone has extra time. Internet-facing firewalls, VPN systems, remote-access platforms, web servers, file-sharing devices, and business applications require rapid attention.
A delay of several weeks may give AI-assisted attackers more than enough time to act.
4. Phishing Messages Are Becoming More Convincing
Poor grammar and awkward wording once helped employees recognize many phishing emails.
Generative AI removes much of that warning.
An attacker can create polished messages tailored to a specific employee, company, industry, or transaction. AI can review social media profiles, websites, press releases, job titles, and public records to make the message more believable.
- A managing partner asking for a wire transfer
- A company owner requesting gift cards
- A vendor changing its banking information
- A client sharing a document
- A Microsoft 365 administrator asking the employee to sign in
- A benefits provider requesting employee information
- A bank asking the recipient to verify a transaction
- An executive sending an urgent message from a mobile device
AI can also improve text messages, fraudulent phone scripts, voice clones, fake identification documents, and video impersonation.
The result is a social-engineering attack that appears professional, relevant, and urgent.
5. Attackers Can Analyze Stolen Data More Efficiently
Stealing information is only part of a data breach. Criminals must determine what the data contains and how it can be used.
A compromised server may hold thousands of documents, email archives, database records, invoices, contracts, tax files, medical information, passwords, and internal communications.
AI can help classify and summarize that material.
- Banking information
- Administrative credentials
- Sensitive client files
- Protected health information
- Attorney-client communications
- Employee tax records
- Insurance documents
- Intellectual property
- Financial statements
- High-value customers
- Information useful for extortion
This can increase the damage caused by a breach because the attacker can locate valuable information faster.
6. Poorly Skilled Attackers Can Still Cause Severe Damage
An inexperienced attacker may make mistakes. That does not make the attacker harmless.
Poorly controlled activity can corrupt data, interrupt business applications, damage servers, expose information, or trigger security systems. An attacker who does not fully understand the environment may cause more disruption than intended.
- Delete the wrong files
- Shut down a production server
- Break a database
- Expose stolen information publicly
- Disable security tools
- Lock legitimate users out
- Interrupt manufacturing or business operations
- Damage backup systems
- Leave additional vulnerabilities behind
Businesses should not assume that an unskilled attacker produces a minor incident. Limited expertise combined with powerful automation can create unpredictable consequences.
7. Cyberattacks Can Be Conducted at Greater Scale
AI excels at repetitive work.
An attacker can use automation to evaluate large numbers of IP addresses, websites, email accounts, passwords, or exposed services. AI can help interpret the results and prioritize the most promising targets.
Instead of focusing on one carefully selected organization, an attacker may test hundreds or thousands of businesses.
This supports a volume-based criminal model. The attacker does not need every attempt to succeed. A few vulnerable organizations may be enough to make the activity profitable.
For Michigan businesses, being relatively small or locally focused does not create protection. Internet-connected systems are visible from anywhere in the world
8. Cybersecurity Guardrails Have Practical Limitations
Leading AI providers implement policies and safeguards intended to prevent harmful use. These controls can stop or restrict certain requests.
However, cybersecurity work is inherently dual-use.
The same technical activities can support either a legitimate security assessment or a criminal intrusion. Network reconnaissance, vulnerability testing, exploit validation, credential analysis, and penetration testing are normal tasks for authorized security professionals.
An AI system may have difficulty determining whether the person requesting assistance has permission from the owner of the targeted system.
- Misrepresent malicious activity as an authorized security test
- Rephrase blocked prompts
- Use older model versions
- Use locally hosted or open-source models
- Combine multiple tools
- Steal authenticated AI-agent installations
- Disable or bypass permission controls
- Use AI for individual steps rather than an entire attack
Businesses cannot rely on AI providers’ safety controls as their primary defense.
The Threat Is Not Limited to Large Companies
Many small-business owners assume sophisticated technology will be used primarily against banks, national corporations, or government agencies.
That assumption is dangerous.
AI makes broad, opportunistic targeting more practical. A local business may be selected simply because it has an outdated system, exposed port, reused password, compromised email account, or employee who responds to a convincing message.
Michigan law firms, healthcare practices, financial professionals, manufacturers, property-management companies, construction businesses, and accounting firms hold information criminals can monetize.
That information may include client records, Social Security numbers, medical data, financial details, legal documents, payment instructions, employee information, and access to larger customers or vendors.
Small businesses may also have fewer security controls, making them easier to compromise and slower to detect an intrusion.
Local risk is still global risk
A business in Warren, Sterling Heights, Macomb County, Oakland County, or Wayne County can be scanned and attacked from anywhere. Geographic size and local market focus do not make internet-facing systems invisible.
How Businesses Can Defend Against AI-Assisted Attackers
The answer is not to purchase a product labeled “AI security” and assume the problem has been solved.
Businesses need layered controls that reduce the opportunities available to both skilled and unskilled attackers.
1. Identify Everything Exposed to the Internet
web applications, remote-access tools, cloud services, email systems, and file-sharing platforms.
Unknown systems cannot be protected effectively.
- Unexpected open ports
- Forgotten web applications
- Exposed administrative interfaces
- Expired certificates
- Misconfigured services
- Vulnerable network devices
- Systems using outdated software
- Publicly accessible storage
- New services appearing without approval
Cybercriminals continuously search for these weaknesses. Businesses should discover them first.
2. Patch Internet-Facing Systems Quickly
Create a formal process for evaluating and installing security updates.
Critical vulnerabilities affecting internet-facing systems may require action within 24 to 72 hours, depending on the severity, availability of exploits, and risk to the business.
- Firewalls
- VPN appliances
- Remote-access software
- Email servers
- Web applications
- File-transfer systems
- Network-attached storage devices
- Cloud management tools
- Public-facing servers
- Business applications with remote portals
Routine monthly patching may not be fast enough for actively exploited vulnerabilities.
3. Require Multifactor Authentication
Multifactor authentication can prevent many account-compromise attempts even when an attacker has obtained a password.
- Microsoft 365 and Google Workspace
- Remote-access tools
- VPN connections
- Administrative accounts
- Accounting platforms
- Cloud management portals
- Password managers
- Backup systems
- Line-of-business applications
- Vendor-access accounts
Whenever possible, use phishing-resistant methods such as security keys, passkeys, or certificate-based authentication. Traditional push notifications should include number matching and protections against repeated approval requests.
4. Remove Unnecessary Administrative Access
Employees should not use administrator accounts for routine work.
Attackers who compromise an account inherit the permissions assigned to that user. Reducing those permissions limits what malware, stolen credentials, and remote intruders can do.
- Separate standard and administrative accounts
- Remove unused local administrators
- Review Microsoft 365 administrator roles
- Restrict vendor access
- Disable dormant accounts
- Use time-limited privileged access where possible
- Monitor administrative logins
- Protect emergency accounts separately
Least privilege remains effective regardless of whether the attacker is human or AI-assisted.
5. Strengthen Email and Payment Verification
Businesses need procedures that assume some fraudulent messages will look legitimate.
- Wire transfers
- Bank-account changes
- Payroll changes
- Gift-card purchases
- Sensitive documents
- Password resets
- New vendor payment instructions
- Unusual login requests
Verification should occur through a known phone number or another trusted communication channel, not by replying to the message.
Email security should also include impersonation protection, malicious-link scanning, attachment analysis, domain-authentication controls, and monitoring for suspicious mailbox rules.
6. Use Endpoint Detection and Application Control
Traditional antivirus alone may not stop an attacker who uses legitimate tools, customized scripts, or newly generated code.
- Credential theft
- Suspicious scripting
- Unauthorized remote-access tools
- Malicious PowerShell activity
- Unexpected encryption
- Security-tool tampering
- Unusual child processes
- Data-staging activity
- Lateral movement
- Persistence mechanisms
Application allowlisting can provide another important layer by restricting programs and scripts that have not been approved.
7. Monitor Systems Continuously
AI-assisted attacks may move faster than a business’s traditional support process.
Waiting until Monday morning to review an alert generated Friday night may be too late.
- Endpoints
- Servers
- Firewalls
- Cloud identities
- Microsoft 365 or Google Workspace
- Email activity
- Backups
- Administrative accounts
- Remote-access systems
- Security logs
A managed security operations capability can investigate alerts, identify related activity, and begin containment when suspicious behavior occurs.
8. Protect Backups From Attackers
Backups should not be continuously accessible through the same accounts and network paths as production systems.
- Encrypted backups
- Offsite or cloud copies
- Immutable storage
- Separate backup credentials
- Multifactor authentication
- Restricted administrative access
- Automated backup monitoring
- Regular restoration testing
A backup is not reliable until the business has successfully restored data from it.
9. Create and Test an Incident-Response Plan
Every business needs a written plan explaining what to do when a cyberattack occurs.
- Who has authority to make decisions
- How to contact IT and cybersecurity support
- How to isolate affected devices
- When to contact cyber insurance
- How to preserve evidence
- How to communicate with employees
- Which legal and regulatory obligations may apply
- How to continue critical operations
- How backups will be restored
- How vendors and clients will be notified
Conduct a tabletop exercise at least annually. Organizations with higher risk, regulatory obligations, or frequent staff changes may benefit from testing every six months.
10. Secure the Business’s Own AI Tools
The reported incident also highlights a separate concern: AI coding agents and their saved sessions can become valuable targets.
- Source code
- Server credentials
- API keys
- Cloud environments
- Deployment systems
- Internal documentation
- Customer information
- Saved command histories
- Business files
Companies that use AI agents should treat them like other privileged business systems.
Protect AI tools with strong authentication, limited permissions, secure credential storage, logging, session controls, data-retention policies, and endpoint protection. Do not paste passwords, private keys, regulated information, or confidential client data into unapproved AI systems.
What Michigan Business Owners Should Do Next
Business owners do not need to become experts in artificial intelligence or offensive cybersecurity.
They do need to confirm that their defenses can withstand faster, more frequent, and more convincing attacks.
Start by answering these questions:
- Do we know which systems are publicly exposed?
- Are our firewalls and remote-access systems fully updated?
- Is multifactor authentication required everywhere it should be?
- Are employees using standard accounts for daily work?
- Can we detect suspicious activity outside normal business hours?
- Are our backups isolated, monitored, and regularly tested?
- Do employees verify financial requests using a second communication method?
- Do we have a documented cyber incident-response plan?
- Has an independent professional tested our security controls?
- Would we know what to do during the first hour of a ransomware attack?
An unanswered question represents a potential business risk.
AI Has Changed the Threat, but Strong Security Still Works
Artificial intelligence gives attackers speed, technical guidance, and scale. It does not make proven cybersecurity controls obsolete.
Updated systems, multifactor authentication, restricted privileges, secure email, application control, endpoint monitoring, protected backups, employee education, and incident-response planning still reduce risk.
The difference is urgency.
Businesses that leave known weaknesses unresolved may now face more attackers capable of finding and exploiting them. Organizations that maintain strong, monitored, and layered defenses will be much harder targets.
Protect Your Business Before an AI-Assisted Attacker Finds the Weakness
Cyber Protect LLC helps small and midsize businesses identify security gaps, strengthen their defenses, monitor for threats, and prepare for cyber incidents.
We provide cybersecurity and managed IT services for organizations throughout Warren, Sterling Heights, Macomb County, Oakland County, Wayne County, and Southeast Michigan. Our experience includes protecting law firms, healthcare practices, financial-services organizations, manufacturers, construction companies, accounting firms, and other growing businesses.
Do not wait for an attacker to show you where your security is weak.
Schedule a cybersecurity and IT risk assessment with Cyber Protect LLC. We will help you identify exposed systems, evaluate your current protections, prioritize practical improvements, and create a plan based on your actual business risk
Frequently Asked Questions
Can unskilled hackers really use AI to attack a business?
Yes. AI can help inexperienced attackers research exposed systems, identify known vulnerabilities, generate commands, troubleshoot failed attempts, create phishing messages, and analyze stolen information. The attacker still needs access, intent, and some ability to direct the process, but AI can substantially increase what that person is capable of doing.
Are AI-powered cyberattacks fully autonomous?
Most current attacks still require human involvement. Attackers typically select targets, provide tools or access, approve actions, and make decisions. However, AI agents can automate long sequences of research, coding, testing, analysis, and technical execution.
Why are small businesses at risk from AI cybercrime?
Small businesses often have valuable information but fewer security resources than large enterprises. Attackers may target them because of outdated systems, exposed remote access, weak passwords, missing MFA, limited monitoring, or untested backups.
Can antivirus stop an AI-generated cyberattack?
Antivirus is one layer of protection, but it is not enough by itself. Businesses also need endpoint detection, multifactor authentication, application control, patch management, secure email, network monitoring, protected backups, employee training, and an incident-response plan.
How quickly should a business install security updates?
Critical updates for internet-facing systems may need to be installed within 24 to 72 hours, particularly when active exploitation has been reported. Other updates should follow a documented, risk-based patch-management schedule.
How can a business prevent AI-generated phishing attacks?
Use advanced email security, multifactor authentication, domain protections, employee training, and mandatory verification procedures for payments, password resets, account changes, and sensitive-data requests. Employees should verify unusual requests using a known phone number or separate trusted channel.
What is the best first step for a small business?
Begin with a professional cybersecurity risk assessment. The assessment should review internet-facing systems, identities, MFA, endpoint protection, email security, backups, patching, administrative access, monitoring, and incident-response readiness.
Does Cyber Protect provide AI cyber-threat protection in Michigan?
Cyber Protect LLC provides managed cybersecurity and IT services to small and midsize businesses throughout Southeast Michigan. Services can include risk assessments, security monitoring, endpoint protection, firewall security, email protection, backup and disaster recovery, vulnerability management, employee education, and incident-response planning.
About the Author
Cheyenne Harden
CEO