Partly. AI has made malware cheaper to build and quicker to change, and that defeats any defence that relies only on recognising files it has seen before. Google's threat intelligence team has documented malware that asks a large language model what to do during an attack, and a dropper that asks an AI model to rewrite its own code, some versions every hour. Anthropic reported a seller with limited coding skill using AI to build ransomware sold for US$400 to US$1,200. A firewall keeps up only if it uses several layers at once: intrusion prevention, sandboxing, behaviour-based and AI-based file analysis, DNS and web filtering, and control over which applications and AI services devices may reach. Even then, the firewall works best alongside endpoint protection and backups. At SourceIT, the Fortinet FortiGate 50G with 1-year Unified Threat Protection (UTP) is S$1,819.20 (listed 30 Oct 2026).
What does AI-assisted malware actually look like?
| Example | What it does | Status | Source |
|---|---|---|---|
| PROMPTSTEAL (also called LAMEHUG) | Queries an open-source LLM through the Hugging Face API to generate Windows commands that collect system data and documents | Used in live operations by Russia's APT28 | Google Threat Intelligence Group, Nov 2025 |
| PROMPTFLUX | Asks Google's Gemini API to rewrite its own VBScript code to evade detection | Experimental | Google, Nov 2025 |
| HONESTCUE | Receives C# source code from an AI API, compiles and runs it in memory, so nothing is written to disk | Seen from Sep 2025 | Google, Feb 2026 |
| "No-code" ransomware | A seller with limited skill used AI to develop ransomware with evasion and anti-recovery features | Sold on forums for US$400 to US$1,200 | Anthropic, Aug 2025 |
| WormGPT 4 | Criminal chatbot that wrote a script to encrypt PDFs plus a ransom note when prompted | US$50 a month or US$220 lifetime | Palo Alto Networks Unit 42, via The Register, Nov 2025 |
From the named threat reports. We have not analysed these samples ourselves.
Why do signatures alone fall behind?
A traditional antivirus signature recognises a known file. Malware that regenerates itself, or never writes a file to disk, has no stable fingerprint. Speed makes it worse: FortiGuard Labs' 2026 threat report puts the time-to-exploit for critical outbreaks at 24 to 48 hours. If the only defence is waiting for a signature, the first day belongs to the attacker.
Signatures still matter, because most malware in circulation is not new. But a firewall that keeps up layers other checks on top of them.
Which firewall layers catch what?
| Attacker technique | FortiGate layer that helps | Licence on FortiGate |
|---|---|---|
| Exploiting a known flaw to get in | Intrusion prevention (IPS), which blocks the exploit even before a patch is installed | ATP, UTP or Enterprise |
| Delivering a new, never-seen file | Cloud sandbox for zero-day detection | ATP and UTP |
| Delivering a new file and needing an instant verdict | AI-based inline malware prevention | Enterprise Protection only |
| Phishing link or malicious download site | Web and DNS filtering | UTP or Enterprise |
| Malware calling home to its controller | Botnet and command-and-control blocking, DNS filtering | Bundles; botnet IP database also with FortiCare |
| Malware calling an AI API at runtime | Application control to restrict which AI services servers and staff can reach | Included with FortiCare |
| Threats hidden in HTTPS | SSL deep inspection, so the engines above can see the content | Built in; costs throughput |
Licence placement from Fortinet's FortiGate security bundles page and FortiGuard ordering guide (May 2026). Note the model limits: Fortinet says Content Disarm and Reconstruction, video filtering and inline CASB are not available on the 30G and 50G series.
No firewall recognises every new file. A good one makes the attacker get past five checks instead of one.
Where does the firewall's job end?
A network firewall sees traffic crossing it. It does not see a file copied from a USB drive, a laptop infected on hotel Wi-Fi, or a malicious script started by a user who was tricked into pasting it. Microsoft's 2025 Digital Defense Report found "ClickFix", which tricks users into running commands themselves, was the most common initial access method its experts saw. That is endpoint territory. Pair the firewall with endpoint detection and response, such as Microsoft Defender for Endpoint, and keep isolated, tested backups so that even successful ransomware is recoverable. Our post Your Business Has Antivirus. So Why Would You Need a FortiGate Firewall? sets out the division of labour.
How to set a FortiGate up to keep pace
- Keep a security bundle active so IPS, antivirus and filtering databases update continuously.
- Run Fortinet's recommended firmware; in October 2026 that is FortiOS 7.6.8 for entry models.
- Enable SSL deep inspection for outbound web traffic, with sensible exemptions such as banking and health sites.
- Apply antivirus and sandboxing to web and email traffic, not just the default profile.
- Use application control to limit which AI services and cloud storage apps are allowed, especially from servers.
- Review blocked-threat logs weekly so you notice when something keeps trying.
For the AI features inside FortiGuard itself, see AI vs AI: Can FortiGate Fight Hackers Using Artificial Intelligence?
Buying in Singapore
SourceIT supplies FortiGate firewalls with UTP bundles and local warranty: the FortiGate 30G (S$1,236 with 1-year UTP), 50G (S$1,819.20), 70G (S$2,630.40) and 90G (S$6,094.80), plus Microsoft Defender for Endpoint P1 (S$46 per year as listed) for the endpoint layer. Prices listed 30 Oct 2026; ask us about Enterprise Protection bundles. Browse the FortiGate firewall collection. For a quotation, use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.
Frequently asked questions
Can AI really write malware? Yes. Google, Anthropic and Palo Alto Networks have documented AI-assisted malware, including code that asks an AI model to rewrite itself and ransomware built by sellers with limited coding skills.
Can a firewall detect AI-generated malware? Often, through layers rather than one signature: intrusion prevention, sandboxing, AI-based file analysis, DNS filtering and blocking of command-and-control traffic. Endpoint protection covers what the firewall cannot see.
Which FortiGuard bundle includes AI-based inline malware prevention? Fortinet lists it in the Enterprise Protection bundle. ATP and UTP include cloud sandbox-based zero-day detection.
Does SSL inspection matter for malware detection? Yes. Most web traffic is encrypted, so without SSL deep inspection the firewall's antivirus and IPS engines cannot see the content of most downloads.
Is a firewall enough against ransomware? No. Combine it with endpoint detection, patching, MFA and isolated, tested backups so an infection can be contained and recovered from.
SourceIT Approved
Researched and verified by the SourceIT team for Singapore buyers.
