{"id":1150632,"date":"2025-10-06T07:06:12","date_gmt":"2025-10-06T14:06:12","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-story&p=1150632"},"modified":"2025-12-15T14:52:59","modified_gmt":"2025-12-15T22:52:59","slug":"the-paraphrase-project-designing-defense-for-an-era-of-synthetic-biology","status":"publish","type":"msr-story","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/story\/the-paraphrase-project-designing-defense-for-an-era-of-synthetic-biology\/","title":{"rendered":"The Paraphrase Project: Designing defense for an era of synthetic biology"},"content":{"rendered":"\n
<\/div><\/span>
\n
\n
<\/div>\n\n\n\n

The Paraphrase Project: Designing defense for an era of synthetic biology<\/h1>\n\n\n\n
<\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n
\n
\n\t\n\t
\n\t\t
\n\t\t\t
<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n
\n
<\/div>\n\n\n\n
\n
<\/div>\n\n\n\n

In the fall of 2023, breakthroughs in generative AI had researchers proclaiming a new era for medicine and healthcare.<\/h2>\n\n\n\n
<\/div>\n\n\n\n
<\/div>\n\n\n\n

In fall 2023, as excitement grew over generative AI\u2019s role in medical breakthroughs, Microsoft\u2019s Eric Horvitz<\/a> couldn\u2019t shake a nagging concern. The promise was undeniable, but his thoughts kept circling back to one hypothetical scenario: that open-source AI tools could, in theory, be used to reengineer toxins capable of evading existing biosecurity software.<\/p>\n\n\n\n

\n\t
\n\t\t
\n\t\t\t\t\t\tPublication<\/span>\n\t\t\tStrengthening nucleic acid biosecurity screening against generative protein design tools<\/span> <\/span><\/a>\t\t\t\t\t<\/div>\n\t<\/article>\n<\/div>\n\n\n\n

That idea became the seed of what would grow into the Paraphrase Project, initially a biosecurity exploration, now the foundation of a landmark paper published October 2 in Science<\/em>.<\/p>\n\n\n\n

\n

\u201cI tend to look out at the AI horizon all the time. And in scanning the horizon, I always reflect on dual use issues,\u201d said Horvitz, Microsoft\u2019s chief scientific officer. \u201cYou can design proteins to take on any structures now. What about known toxins?\u201d<\/p>\n<\/blockquote>\n\n\n\n

This question is what led Horvitz to spearhead the first biological \u201czero day\u201d: a term borrowed from cybersecurity to describe a previously unknown vulnerability in a computer system. In the realm of biosecurity, the analogy isn\u2019t perfect. This was about preemptively identifying cracks in the system before they could be exploited, not patching an active breach. But it was the closest analogy to what they were doing\u2014probing for hidden vulnerabilities and imagining how future misuse might unfold before it ever has the chance. <\/p>\n\n\n\n

At its core, the Paraphrase Project<\/a> calls for a shift in how we evaluate biological sequences, moving beyond surface-level identity checks to a deeper, semantic understanding of what proteins actually do. This approach could dramatically improve the safety and reliability of AI-driven biotechnologies, especially as synthetic biology becomes more accessible, Horvitz said. <\/p>\n\n\n\n

\n