Silicon Valley looks at microscopes. How AI became the top intelligence radar

Anthropic published a study in which he boasts of blocking abroad military-related researchers utilizing Claude's model to work on a dangerous virus.

The communicative sounds like a ready-made introduction to the global pandemic film, but the most interesting thing about it is simply a completely different theme. Turns out, in our pursuit of intelligent assistants, we created a strategy that acts like a global intelligence sensor. AI looks at laboratory logs all over the planet today, and scientists themselves tell machines their top secrets.

Blurring the border between vaccine and weapon

Incident reported Anthropic, it was not that a abroad scientist asked artificial intelligence to rewrite a simple grant application. The study leaves no illusions: Claude besides helped him to prepare another stages of the investigation project. The model generated hypotheses, designed Gain-of-function experiments (amplifying the chikungunya virus), selected variants with greater virulence, determined dosage, statistic, and even emergency strategies.

What's so scary about that? Not that the device designed possible biological weapons (with emphasis on “potential”). The biggest problem – to which Anthropic admits himself – is that present it is impossible to separate with 100% certainty the intentions of the researcher. Biology is the discipline of dual use. A set of the same instructions and genetic modifications can be utilized to pre-create a life-saving vaccine as well as to build sophisticated biological weapons.

Random Honeypot and the end of conventional spies

And that's where we get to the most interesting point. Models of artificial intelligence, advertised as universal tools for scientists, have become a giant blank for information. Researchers – including those associated with the military who bypass regional blockades through intermediaries and VPNs – voluntarily introduce into language models the assumptions of their most secret experiments.

Historically, tracking this kind of activity required immense investments. We had to send intelligence officers, trust on UN inspectors, analyse recovered equipment or look for gaps in technological publications. present Anthropic sees it all in logs of your servers. erstwhile analysing usage patterns and associated accounts, the corp can monitor activities that previously were reserved exclusively for the intelligence services of the most powerful states.

Cloud AI is simply a mystery vacuum cleaner. Samsung escapes from Silicon Valley and locks Mistrala in the basement

We don't gotta add cynical marketing theories to that. The facts defend themselves. The latest language models have reached the threshold at which they truly aid in complex biological engineering. But their most interesting function is not that they come up with fresh science, but that they are silently looking at abroad powers on their hands. We've built an AI for scientists, and we've been given a strategy where military experts themselves give Americans on a tray of what they're presently working on.

If article Silicon Valley looks at microscopes. How AI became the top intelligence radar does not look right in your RSS reader, then see it on iMagazine.