AI-Powered Cybercrime Is Becoming an SME Problem
Artificial intelligence is changing how businesses work, but it is also changing how criminals attack them. A new threat report from Anthropic says its team identified and disrupted malicious operations involving Claude between December 2025 and August 2026, covering cyber operations, scams, surveillance, influence activity, biological misuse, conventional weapons development and illicit model distillation. Read the full Anthropic report.
For a Malaysian SME, the important message is not that every attacker now has advanced artificial intelligence. It is that ordinary criminals may be able to work faster, target more businesses and produce more convincing messages with fewer technical skills. Your business may have only 10 or 20 employees, but you still hold customer information, payment details, supplier records, employee data and access to online banking or cloud systems.
The report describes a shift from AI acting as a simple chatbot assistant to AI helping coordinate parts of an attack. That means your risk is no longer limited to someone receiving a poorly written phishing email. A criminal operation may now involve automated research, personalised messages, technical probing and repeated attempts to bypass security controls.
What Happened
Anthropic’s Threat Intelligence team reported that suspected state-backed groups, financially motivated criminals, commercial spyware vendors, propaganda institutions and politically motivated individuals attempted to misuse Claude. The cases involved Claude Haiku, Sonnet and Opus models; the report states that no misuse cases involved Claude Fable or Mythos-class models, except for one illicit distillation case. These details are available in Anthropic’s September 2026 announcement.
In the cyber cases, the company said AI supported activities across the cyber kill chain, including reconnaissance, tool development, exploitation, data processing and possible data exfiltration. The report also said that several operations used multi-agent frameworks, with humans setting targets and reviewing results. This is significant because the attacker does not necessarily need to be a highly skilled expert if AI can handle more of the repetitive and technical work.
One case, identified by Anthropic as GTG-20006, involved an actor linked in the report to Russian state-nexus espionage activity. The operation used an AI-assisted workflow that could automatically rebuild and redeploy tools after security products detected them. The report also discusses GTG-50014 and other cases to illustrate how AI can increase the speed, scale and depth of malicious activity. See the source report for the case studies and methodology.
“Sophistication has stopped being a reliable signal of who is behind an operation,” Anthropic wrote when describing how AI is narrowing the gap between well-resourced attackers and less experienced operators. Read the statement in the original report.
The report also covered a network of fake dating apps allegedly designed to defraud users, surveillance systems intended to identify and monitor dissidents, influence activity and other forms of misuse. These examples show that AI security is not limited to hacking. It also affects trust, identity, communications and the authenticity of online interactions.
Why This Matters for Malaysian SMEs
Malaysian SMEs are attractive targets because they often combine valuable information with limited internal security resources. A trading company may have supplier bank details and customer invoices. A clinic may store appointment and health information. A construction firm may hold tender documents and project plans. A retailer may depend on payment systems, delivery platforms and social media accounts. If one employee’s account is compromised, the attacker may be able to move into several connected services.
AI-assisted phishing is particularly relevant. A message can be written in clearer Bahasa Malaysia or English, reflect the sender’s industry and imitate the tone of a real supplier. It may refer to an actual invoice, delivery, tender or renewal date gathered from public information or a compromised mailbox. The danger is not only grammatical accuracy. The message may arrive at the right time and request a realistic action, such as changing a bank account, opening a document or approving a payment.
Your business may also face attacks through third parties. If your accountant, logistics provider, software vendor or supplier has a compromised account, criminals may use that trusted relationship to contact you. This is why checking the sender’s identity alone is no longer enough. You need a separate verification process for sensitive requests.
| Business risk | What an attacker may attempt | Practical control for you |
|---|---|---|
| Payment fraud | Change a supplier’s bank account or create an urgent transfer request | Verify changes using a known phone number and require two-person approval |
| Account takeover | Steal email, cloud or social media credentials | Use multifactor authentication and remove unused accounts |
| Data theft | Copy customer, employee, quotation or supplier records | Restrict access by role and review cloud sharing permissions |
| Malware delivery | Send a fake invoice, quotation or delivery document | Train staff to report unexpected attachments and verify requests |
| Reputation damage | Post from a hijacked social account or send fraudulent messages | Use separate administrator accounts and maintain an incident plan |
You should also consider how your own employees use AI tools. Staff may paste customer lists, contracts, source code or internal financial information into an external service without understanding the data-handling implications. Establish a written policy that explains which information may be entered into AI tools, which tools are approved and who is responsible for checking generated content.
What You Can Do This Month
Start with your most valuable accounts. List your business email, cloud storage, accounting platform, payroll system, website administrator, online marketplace accounts and social media pages. Confirm who has access, activate multifactor authentication wherever available and remove former employees or unused administrator accounts.
Next, create a payment-verification rule. Any request to change bank details, make an unusual transfer or bypass normal approval must be verified through a separate communication channel. Do not use the telephone number or reply address included in the suspicious message. Use the supplier information already stored in your records.
Then, run a short staff exercise. Send a harmless simulated example or discuss a realistic scenario involving a fake supplier, urgent director request or delivery document. Teach employees to pause, report and verify rather than blame themselves for making a mistake. A reporting culture gives you more time to contain an incident.
Keep backups of essential files and test whether you can restore them. A backup that has never been tested may not be useful during a ransomware or account-compromise incident. Record who should contact your IT provider, bank, insurer, platform provider and relevant authorities if something goes wrong.
The Bigger Picture
Anthropic’s report suggests that defensive teams can no longer rely only on recognising highly sophisticated attack tools. If AI helps different types of criminals conduct activities that previously required specialist teams, small businesses need layered controls rather than one protective product. The relevant measures include strong authentication, restricted permissions, verified payments, reliable backups, staff awareness and clear response procedures. These recommendations respond to the trends described in the Anthropic threat intelligence report.
This does not mean you should avoid useful AI tools. Automation can help you draft documents, answer routine enquiries, organise records and improve productivity. The safer approach is to adopt AI with clear boundaries. Keep sensitive information out of unapproved tools, require human review for important decisions and log who can access business data.
The practical lesson for a Malaysian SME is straightforward: assume that fraudulent messages will become more convincing, attacks will become more persistent and technical expertise will be less important for criminals. Build processes that do not depend on everyone spotting every scam. When payment changes require verification, administrator access is limited and suspicious activity is reported quickly, your business becomes harder to exploit—even as attackers gain better tools.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →