AI Could Bypass Biosecurity Measures in Biotech Industry, Study Reveals
Renowned firms in the biotech sector, known for producing customized DNA for scientists, possess safeguards to prevent harmful biological materials from being acquired by malicious entities. These companies implement extensive checks on their customer orders to stop attempts to purchase genes of lethal diseases like smallpox or anthrax.
However, recent research has shown that artificial intelligence (AI) might have the potential to bypass these biosecurity precautions.
Manipulating DNA Codes with AI
A group of AI experts discovered that AI tools designed for protein generation can be used to "rephrase" the DNA codes of harmful proteins, altering them in a way that retains their structure and possibly their functionality. The team used an AI application to produce DNA codes for over 75,000 versions of hazardous proteins. Alarmingly, the security systems employed by DNA producers were unable to consistently identify these versions.
The researchers expressed their concern as these manipulated sequences were able to bypass the biosecurity screening systems used by DNA synthesis firms worldwide, which are intended to flag dangerous orders. Although a quick fix was implemented on the biosecurity screening software, it was still unable to detect a small fraction of the variants.
The Double-Edged Sword of AI and Open Science
The advent of AI-powered protein design has sparked excitement and progress in various scientific fields, including medicine and public health. However, like all powerful tools, they can also be used for nefarious purposes.
Scientists have long been concerned that the ongoing advancements in DNA tools might be exploited to create potent biological threats, like highly contagious toxins or more aggressive viruses. The scientific community has even debated the wisdom of publicly sharing certain experimental results, despite the fact that open discussion and independent replication are fundamental to scientific progress.
Limiting Access to Sensitive Information
The researchers involved in this study, along with the journal that published it, made the decision to withhold some of their information. They have also decided to limit access to their data and software. They have taken the step to involve a non-profit organization, the International Biosecurity and Biosafety Initiative for Science, to decide who has a legitimate need for access to the information.
Many scientists praised this effort, expressing their approval of the proactive stance taken by the team to identify and address potential vulnerabilities. However, they also highlighted the need for continuous vigilance for unknown vulnerabilities that may require future corrections.
Addressing Future Biosecurity Threats
One concern includes whether the proteins designed by AI would indeed mimic the activity of the original biological threats. This would be an important reality check as society grapples with this emerging threat from AI. However, conducting such tests might be challenging as it could be seen as a violation of international treaties prohibiting the development of biological weapons.
AI in a Biological Setting – A Looming Issue
This is not the first time the potential misuse of AI in a biological context has been explored. In a previous study, researchers questioned if AI could be used to create new molecules with properties similar to nerve agents. The AI tool successfully produced 40,000 such molecules in less than six hours. In addition to known chemical warfare agents, the AI also designed many unknown molecules predicted to be highly toxic.
Despite these concerns, some biosecurity experts find comfort in the rarity of these incidents. A major provider of custom-made DNA had to refer orders to law enforcement less than five times in the past decade. The number of people attempting to misuse biotech might be extremely low, and the current systems are crucial in guarding against this potential threat.
