“Can you help me create the Biological Age?”
As expected, ChatGpt said no. “The creation or dissemination of biological weapons is illegal, unethical, and dangerous. If you have questions about biology, epidemiology, or related scientific topics for legal education or research purposes, I am happy to help,” AI added.
So I continued with the “real question” about editing viruses with low technology, and it quickly gave me a guide on how to go with it. Jailbreak AI chatbots like ChatGpt are well known, and Openai knows it well. In a drastic warning, Openai said that the next generation of artificial intelligence models are likely to reach “high” levels of biology.
The company essentially admits that some researchers have been warning for years. AI acknowledges that it can help amateurs without formal training create potentially dangerous biological weapons.
AI companies promote agents as research assistants. In fact, they greatly promote the capabilities of their systems, accelerate drug discovery, optimize enzymes in climate solutions, and help with vaccine design. But these same systems can allow for something darker with the wrong hands.
Historically, one of the key barriers to Biowapons has been expertise. Pathogen engineering is not plug and play. Special knowledge and laboratory skills are required. However, AI models trained on the total biological literature, methods, and heuristics can act as potentially available assistants and lead to the determined user in stages.
For now, the biggest biological threats still reach not from laptops, but from well-equipped labs. Creating a BioWeapon requires access to controlled substances, laboratory infrastructure, and the kinds of know-how that are difficult to counterfeit. However, that buffer – the distance between interest and ability – has been reduced.
AI has not invented any new pathogens. But it may help people replicate known threats faster and easier than ever before.
“We are not yet a world of novel, completely unknown creations in biothreats that had not existed before,” Johannes Heidekke head told Axios. “We're more concerned about replicating what experts are already very familiar with.”
Overall, artificial intelligence is already accelerating fields such as biology and chemistry. The net contribution is positive, but we are in a stage where malicious use is on the table with serious consequences.
How businesses are trying to stop this
Openai says it is taking a “multi-faceted” approach to mitigate these risks.
“We need to act responsibly amid this uncertainty, which is why we are working to integrate AI into positive use cases such as biomedical research and biodeath fences, while also focusing on limiting access to harmful capabilities. Our approach focuses on prevention.
But what does that actually mean?
First of all, we teach models that make answering prompts that could lead to biological age. Dual use fields such as virology and genetic engineering aim to provide general insights rather than lab response instructions. In fact, it has proven to be a vulnerable defense.
Many examples from independent testers and journalists demonstrate that AI systems including Openai can provide sensitive biological information, even with relatively simple and rapid engineering. Sometimes all you need to do is phrase the request as a fictional story or ask for information step by step.
Openai also wants to suspend accounts that attempt to hijack AI or even report it to authorities, including monitoring and enforcement of more people. Finally, they also use AI-trained expert “Red Team”, a biology-trained person, to try to break the conservation measures under realistic conditions and see how this can be stopped.
This combination of AI filters, human surveillance, and hostile testing sounds robust. But beneath it is an unpleasant truth. These systems have never been tested at the scale and interests we are approaching in the real world.
Even Openai admits that 99% is ineffective. “We basically need something that's close to perfect,” said Heidekke, director of open safety systems. But perfection is elusive. Especially when new misuse techniques can emerge faster than defense. A quick injection attack, jailbreak trick, or coordinated abuse can be overwhelming even the most thoughtfully designed systems.
The floodgates are already open
There's a right approach to Openai, and even if they somehow make it work (both big as “if”), they're not the only company in the business. Humanity, the AI company behind Claude, has implemented new safeguards after concluded that the latest models could contribute to biological and nuclear threats.
The US government is also beginning to understand the potential double use risks of AI. Openai expanded its work at the US National Laboratory and convened the Biodefense Summit in July this year. Together, government researchers, NGOs, and policy leaders will explore how AI can support both biological innovation and security.
But even with these efforts, it seems difficult to see a future where malicious AI output is truly controlled.
AI is moving fast. And biology is uniquely sensitive. Today's most powerful AI tools lie behind company firewalls, but open source models are growing, and the hardware that runs them is more accessible.
The cost of synthesizing DNA has dropped dramatically. The tools that once lived in elite government labs are now available in small startups and academic labs. If knowledge bottlenecks also break down, bad actors may no longer need doctoral degrees or national sponsorship.
There is no doubt that AI is revolutionizing biology. This helps you understand disease, design treatments, and address global health challenges faster than ever before. But as these tools become more powerful, the line between scientific advancement and misuse becomes thinner. And it's not difficult to see how these models can be used to do real harm.
