“The patch is incomplete, and the state of art is changing. But this is not one and done. This is the beginning of even greater testing, ”says Adam Klor, director of research and development of Technology in Integrated DNA Technologies, a large DNA manufacturer, which is a co-author of Microsoft. “We are something like a race of weapons.”
To make sure that no one uses the study, the researchers say, they do not reveal some of their codes and do not reveal what toxic proteins they asked for alterations. Nevertheless, some dangerous proteins are well known as Ricin-Jad, found in castor beans, and infectious prions that cause a disease of crazy skin.
“This conclusion, combined with rapid achievements in biological modeling with AI support, demonstrates a clear and urgent need for improved procedures for screening for nucleic acid synthesis, combined with a reliable coercion and verification mechanism,” says Dean Ball, an employee on the basics of American innovation, thought tank in San France.
Bolla notes that the US government is already considering checking the order of DNA of the key line of security. Last May, In the enforcement position on the safety of biological researchPresident Trump called for a general updating of this system, although so far the White House has not released new recommendations.
Others doubt that commercial DNA synthesis is the best point of protection against bad actors. Michael Cohen, AI security researcher at the University of California in Berkeley, believes that there will always be ways to disguise sequences, and that Microsoft could be more diligent.
“The task seems weak, and their corrected tools greatly fail,” says Cohen. “It seems that there is an unwillingness to admit that in the near future we will have to retreat from this proposed throttle point, so we must begin to look for the land that we can really keep.”
Cohen says that bios safety should probably be built into the artificial intelligence systems themselves – as directly, or through control over what information they give.
But Clore says that monitoring of genet synthesis is still a practical approach to the detection of biotropes, since several companies that work closely with the government prevail in the production of DNA in the United States. On the contrary, the technology used to create and training of AI models is more common. “You cannot put this gin back in a bottle,” says Clore. “If you have resources to try to deceive us in creating a sequence of DNA, you can probably teach a large language model.”
