International researchers have discovered a serious flaw in biosecurity software that was supposed to block access to dangerous genetic materials. The vulnerability showed that algorithms are not always able to detect artificially modified toxic proteins. This is reported by the Financial Times with reference to research data in the journal Science, writes UNN.
Details
In the Science article, scientists called this the first "zero-day" event in biosecurity – a term that in cybersecurity means a weak link unknown to developers.
AI-based protein design is one of the most exciting fields of science… however, these same tools can be misused
During tests, the team created more than 75,000 variants of dangerous proteins. Some of them were able to pass undetected even after the software was fixed. According to experts, about 3% of the variations retained potentially harmful properties.
This work provides practical and timely protection that enhances current DNA synthesis screening
Scientists emphasize that the development of AI can make the creation of biological weapons cheaper and easier.
We as a society really need to take this seriously now