Microsoft’s research team claims to use artificial ielligence to discover a “zero day” vulnerability in biomass systems designed to preve DNA abuse.
These screening systems are designed to preve the purchase of genetic sequences that can be used to produce deadly pesticides or pathogens. But now researchers led by Microsoft’s top scieist Eric Horwitz say they have found a way to bypass these protections. This method was previously unknown to defenders. The results of these scieists are published in the journal Science and are publicly available.
Artificial Ielligence has now found its way to biology
Horitz and his colleagues focused on productive artificial ielligence algorithms that offer new forms of proteins. These types of applications help search for new drugs by wealthy startups such as Generate Biomedicines and Isomorphic Labs.
But the problem arises from where such systems have dual use, meaning they can use their educational data both to produce useful molecules and for harmful molecules.
Microsoft says a “rejection” test on the dual poteial of artificial ielligence has begun to determine if artificial ielligence protein design can help biologists build harmful proteins.
The tool that Microsoft attacked is what is known as a biological screening software. In order to produce a protein, researchers usually have to order the corresponding DNA sequence from a commercial seller that they can then eer io a cell. Those sellers use screening software to compare the input orders with known toxins or pathogens. Careful matching leads to the warning.
According to the researchers, the experime was completely digital and they did not produce any toxic protein. This was done to preve any perception that the company is developing biological weapons.
Microsoft says it had informed the US governme and software creators before the results are released. Of course, some molecules designed with artificial ielligence can still escape the ideification.
