The scenario of using artificial intelligence for the creation of biological weapons is one of the most serious and debated risks. Unlike the nuclear threat, which requires huge industrial capacities, biological weapons primarily require information and access to laboratories. AI can drastically reduce this barrier.

Key Risk: Lowering the Entry Barrier

The main threat is not that AI itself wants to kill someone, but that it can make the creation of biological weapons accessible to a small group of people or even individuals. If before this required years of training and access to complex materials, now AI can take on a significant part of the intellectual work.

In 2026, the company Antropic published a report describing five real cases where researchers from different countries used their model for potentially dangerous biological work. In particular, one user planned to use AI to search for mutations in the chikungunya virus that increase its ability to spread and evade immunity. Another researcher was working on adapting highly pathogenic avian influenza to mammals, which could enhance its pandemic potential.

When the model refused to respond to dangerous requests, users found bypasses. In one case, requests were redirected to other models with less strict settings, and access to the system was obtained through third-party platforms for circumventing geographical restrictions.

How It Could Happen: A Chain of Events

Technically, the scenario of a catastrophe could look like this.

1. Design generation. AI, trained on vast amounts of biological data, designs the genome of a new pathogen. There are already experiments where AI created functional viruses from scratch, proving the possibility of generating viable genetic sequences.
2. Property optimization. The model can suggest which specific mutations to introduce to make the virus more contagious or resistant to existing vaccines and medications.
3. Synthesis and assembly. The design is passed to the laboratory for DNA synthesis. This is the most complex physical stage, requiring equipment and materials, but AI can help with logistics planning and even in masking orders for the synthesis of dangerous sequences.
4. Spread. The ready-made pathogen is sprayed or otherwise released into the target population.

Why It's Challenging But Not Impossible

Significant obstacles stand in the way of realizing this scenario.

Physical limitations: AI cannot create a virus in a test tube on its own. For this, real laboratories, equipment, and materials are needed. Experts note that creating a virus dangerous to humans, not bacteria, requires huge efforts, time, and money.

Unpredictability of biology: biological systems are chaotic. The designed pathogen may mutate, lose lethality, or, worse, return to the creator. It is due to this unpredictability that most countries have abandoned offensive biological programs.

Defensive mechanisms: companies developing AI are already implementing filters to block dangerous requests. However, as the report showed, these filters can be bypassed, and they are not always effective against determined wrongdoers.

What Can Be Done Against It

Experts agree that multilevel protection measures are needed. The following approaches are already being discussed.

• Compulsory screening of gene synthesis. Companies selling synthesized DNA should check if customers are ordering dangerous sequences.
• International regulation. The existing Biological Weapons Convention is outdated and does not take into account the challenges of AI. New treaties are needed, covering both AI models and biological materials.
• Early warning. Creating systems for rapid detection of outbreaks of new diseases, which can serve as a deterrent.

This way, AI does not create biological weapons on its own, but it is a powerful accelerator and advisor for those who have the corresponding intentions and access to laboratories. The risk lies in the fact that the entry barrier to this field is becoming alarmingly low.

Permanent link to this publication: