Experts tried to get AI to create malicious security threats - but what it did next was a surprise even to them

1. Pro
2. Security

Experts tried to get AI to create malicious security threats - but what it did next was a surprise even to them News By Efosa Udinmwen published 24 November 2025

LLMs generate malicious code, but they fail to run consistently, report finds

Comments (0) ()

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

(Image credit: Shutterstock)

• Report finds LLM-generated malware still fails under basic testing in real-world environments
• GPT-3.5 produced malicious scripts instantly, exposing major safety inconsistencies
• Improved guardrails in GPT-5 changed outputs into safer non-malicious alternatives

Despite growing fear around weaponized LLMs, new experiments have revealed the potential for malicious output is far from dependable.

Researchers from Netskope tested whether modern language models could support the next wave of autonomous cyberattacks, aiming to determine if these systems could generate working malicious code without relying on hardcoded logic.

• Amazon Black Friday deals are live: here are our picks!

The experiment focused on core capabilities linked to evasion, exploitation, and operational reliability - and came up with some surprising results.

You may like

• Can top AI tools be bullied into malicious work? ChatGPT, Gemini, and more are put to the test, and the results are actually genuinely surprising
• Hackers are using GPT-4 to build a virtual assistant - here's what we know
• One in five security breaches now thought to be caused by AI-written code

Reliability problems in real environments

The first stage involved convincing GPT-3.5-Turbo and GPT-4 to produce Python scripts that attempted process injection and the termination of security tools.

GPT-3.5-Turbo immediately produced the requested output, while GPT-4 refused until a simple persona prompt lowered its guard.

The test showed that bypassing safeguards remains possible, even as models add more restrictions.

After confirming that code generation was technically possible, the team turned to operational testing - asking both models to build scripts designed to detect virtual machines and respond accordingly.

Are you a pro? Subscribe to our newsletter

Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!

Contact me with news and offers from other Future brandsReceive email from us on behalf of our trusted partners or sponsorsBy submitting your information you agree to the Terms & Conditions and Privacy Policy and are aged 16 or over.

These scripts were then tested on VMware Workstation, an AWS Workspace VDI, and a standard physical machine, but frequently crashed, misidentified environments, or failed to run consistently.

In physical hosts, the logic performed well, but the same scripts collapsed inside cloud-based virtual spaces.

These findings undercut the idea that AI tools can immediately support automated malware capable of adapting to diverse systems without human intervention.

You may like

• Can top AI tools be bullied into malicious work? ChatGPT, Gemini, and more are put to the test, and the results are actually genuinely surprising
• Hackers are using GPT-4 to build a virtual assistant - here's what we know
• One in five security breaches now thought to be caused by AI-written code

The limitations also reinforced the value of traditional defenses, such as a firewall or an antivirus, since unreliable code is less capable of bypassing them.

On GPT-5, Netskope observed major improvements in code quality, especially in cloud environments where older models struggled.

However, the improved guardrails created new difficulties for anyone attempting malicious use, as the model no longer refused requests, but it redirected outputs toward safer functions, which made the resulting code unusable for multi-step attacks.

The team had to employ more complex prompts and still received outputs that contradicted the requested behavior.

This shift suggests that higher reliability comes with stronger built-in controls, as the tests show large models can generate harmful logic in controlled settings, but the code remains inconsistent and often ineffective.

Fully autonomous attacks are not emerging today, and real-world incidents still require human oversight.

The possibility remains that future systems will close reliability gaps faster than guardrails can compensate, especially as malware developers experiment.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!

And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.

Efosa UdinmwenFreelance Journalist

Efosa has been writing about technology for over 7 years, initially driven by curiosity but now fueled by a strong passion for the field. He holds both a Master's and a PhD in sciences, which provided him with a solid foundation in analytical thinking.

You must confirm your public display name before commenting

Please logout and then login again, you will then be prompted to enter your display name.

Logout Read more Can top AI tools be bullied into malicious work? ChatGPT, Gemini, and more are put to the test, and the results are actually genuinely surprising    Hackers are using GPT-4 to build a virtual assistant - here's what we know    One in five security breaches now thought to be caused by AI-written code    Researchers claim ChatGPT has a whole host of worrying security flaws - here's what they found    AI is creating code faster - but this also means more potential security issues    How many malicious docs does it take to poison an LLM? Far fewer than you might think, Anthropic warns    Latest in Security Black Friday shopping scams are on the rise - experts warn many new domains could be dodgy, here's what to look for    Windows Server flaw targeted by hackers to spread malware - here's what we know    Cox Enterprises hit by Oracle data breach - but it won't name who carried out the attack    Iberia tells customers it was hit by a major security breach    Google security experts say Gainsight hacks may have left hundreds of companies affected    Perplexity responds to Comet browser vulnerability claims, argues "fake news"    Latest in News Sam Altman and Jony Ive AI device is now in its prototype phase and its 'vibe' is defined    Where to watch Blossoms Shanghai online — for *FREE*    Fed up with sluggish folders in Windows 11? Microsoft says it's fixing this    Disney's new Olaf robot is so real, it'll give you chills    ChatGPT’s new Shopping Research tool compares products for you    What is the release date for Pluribus episode 5 on Apple TV?    LATEST ARTICLES

1. 1Sam Altman and Jony Ive AI device is now in its prototype phase and its 'vibe' is defined
2. 2What is the release date for Pluribus episode 5 on Apple TV?
3. 3Black Friday shopping scams are on the rise - experts warn many new domains could be dodgy, here's what to look for
4. 4Dust Bunny actor says it's about more than the monsters under the bed – it's about the world trying to convince us that we're alone
5. 5I tested the entry-level turntable that everyone raves about, and I totally get it