# AI could help terrorists build bombs and bioweapons, report warns > Source: > Published: 2026-08-24 09:34:09+00:00 # AI could help terrorists build bombs and bioweapons, report warns **READ MORE:**[Can you distinguish between real and AI–generated stories?](/sciencetech/article-16027207/spot-fake-test-distinguish-real-AI-stories.html)**See more Daily Mail on Google -**[save us as a Preferred Source](https://google.com/preferences/source?q=dailymail.com) [AI](/sciencetech/ai/index.html) could help terrorist groups build sophisticated biological and chemical weapons, a report has warned. Security researchers warn that popular 'off–the–shelf' AI models can easily be modified to bypass any safety measures. According to the report, [published by research group Crimson Flare](https://crimsonflare.com/ata-report/), this allows an AI to 'walk unsophisticated malicious actors through the specific steps to create explosives, or chemical and biological weapons'. That means so–called 'bad actors' with even a basic grasp of science could soon [arm themselves with incredibly deadly weapons](/sciencetech/article-15126183/rapture-Apocalypse-really-look-like-expert.html) such as anthrax. The report warns of a new method, informally known as 'abliteration', that allows the safety measures in most AI models to be 'disabled in minutes'. These 'abliterated' models could then be uploaded to the internet, where anyone could download and use them for their own purposes. While some models only give basic or unhelpful advice, the researchers believe others give details that make their proliferation a serious national security risk. The report adds: 'Capable, uncensored, open–weight large language models (LLMs) running offline on consumer hardware, represent a serious and growing national security threat.' AI could help terrorists groups build sophisticate biological and chemical weapons, a report has warned. Pictured: Officials take part in a biological weapon attack training exercise in Seol, South Korea The report looks at open–weight LLMs which use the same basic architecture as chatbots like OpenAI's ChatGPT, but can be downloaded and run freely on anyone's computer. For example, Meta recently released its latest open–weight model, Muse Glimmer, and already has several other 'Llama' models available to download. Other popular alternatives include the Chinese Qwen3.8–27b model, released by Qwen, an AI lab owned by the tech giant Alibaba. Normally, these models have certain safety restrictions built into their architecture that prevent users from making certain requests. If a user asks for something restricted, like advice on building a bomb, the AI will return a message saying 'this content can't be shown' or a similar warning. However, some bad actors have developed a technique that finds and removes parts of the model that normally keep it within safe guardrails. Crimson Flare says that these models could, in theory, help malicious actors with any activities that a typical model would refuse, as their 'refusal mechanism' has been removed. The tools to carry out this 'abliteration' are widely available online and can be implemented with relatively basic coding knowledge. Terrorists could hijack popular open–source AIs, such as the Chinese Qwen3.8–27b model, using a technique called abliteration More worryingly, dozens of pre–abliterated models can easily be found and downloaded with a simple search. Within minutes, the Daily Mail was able to find several 'uncensored' versions of models, including Qwen3.8–27b. The creator of one model bragged that their creation would answer on 'tools, chemistry, exploit–shaped code, violence, sexuality, ideologies the publisher trained away from, content that may be illegal where you are'. They added: 'The model does not decide whether to comply. You do.' The report points out that the risk of a terror attack involving chemical and biological weaponry is generally very low due to the technical barriers to producing these tools. The knowledge and skills involved in making a bioweapon are typically restricted to institutional scientists. But powerful AIs without safety restrictions suddenly [put this knowledge in the hands of anyone who wants it.](/sciencetech/article-15169093/AI-new-VIRUSES-sparking-fears-bioweapon.html) Professor Alastair Hay, a British toxicologist and expert on chemical warfare, told the Daily Mail: 'AI lowers the technical threshold required to make dangerous substances. Researchers discovered that AI models could have their safety restrictions removed to give advise on how to conduct biological and chemical weapon attacks. Pictured: A sample response with dangerous information redacted 'In the past we were reassured that the technical knowledge required just to protect the maker was such that those with poor training were a risk to themselves if they embarked on making really hazardous chemicals. Now, with cloud laboratories, much might be farmed off.' After seeing results provided by one of the AI models, Professor Hay told researchers: 'It's almost like a recipe: it gives the concentrations to use. That's pretty good, really. That's almost student level lab work.' Likewise, Zain Ul–Haq, a former Digital Forensics Lead at Lancashire Constabulary, said that AI was a 'one–stop shop' that 'does that all for you … in your own home'. More worryingly, these models can now be run entirely from a user's personal computer, rather than a company's cloud servers. Since the AI is being run entirely on someone's personal computer, MI5 and GCHQ have no way of monitoring use history or keeping an eye out for suspicious behaviour. Even as recently as 2022, no households had the computing power to run large, harm–enabling LLMs offline. Now, models have become so efficient and computing power so easily available, that 87 per cent of households could run their own offline AI with the tech they have now. That means millions of people currently have the ability to access all the information they need to start producing explosives, chemical weapons and deadly pathogens.