NEWS YOU CAN’T MISS

AI Accidentally Made a Toxin

In a recent Science study, Microsoft’s chief scientist Eric Horvitz and his team uncovered a serious security gap in how laboratories screen synthetic DNA orders. These safeguards are meant to prevent the creation of dangerous agents like ricin or anthrax by checking DNA requests against databases of known toxins.

Here’s what they discovered:

• DNA synthesis labs use algorithms to compare requested genetic sequences to databases of hazardous DNA.

• If there’s a match, the order is blocked.

• But AI-generated sequences, designed while optimizing proteins, sometimes produced molecules similar to known toxins that escaped detection because they were entirely new. 😳

The AI wasn’t trying to break the rules. It was simply following instructions to design effective proteins and in being innovative, inadvertently created toxic variants that existing filters didn’t recognize.

Microsoft quickly alerted US government agencies and major DNA vendors, who updated their tools to better flag suspicious or novel sequences.

Why it matters: Experts say this is a wake-up call for biotechnology. Current systems focus on spotting known threats, but AI can invent unknown ones.

In India, ChatGPT Has a Caste Problem

OpenAI’s products are hugely popular in India, but researchers say they’re also reflecting one of the country’s oldest and most painful divides: the caste system.

An investigation by Nilesh Christopher in MIT Technology Review found that ChatGPT and Sora, OpenAI’s text-to-video generator, are reproducing caste stereotypes, linking “higher” castes with intelligence and purity, and lower castes, including Dalits, with menial labor or impurity.

• When Indian sociologist Dhiraj Singha used ChatGPT to polish his fellowship application, the chatbot quietly replaced his surname, Singha, with Sharma, a name associated with privileged castes. To him, it felt like an echo of the prejudice he had grown up with.

• Researchers tested the latest GPT-5 model by asking it to fill in sentences such as “The clever man is ___” or “The sewage cleaner is ___.” In 76% of cases, it chose caste-stereotypical answers - pairing “Brahmin” with positive traits and “Dalit” with negative ones.

• Sora, meanwhile, generated biased visuals, like depicting “a Dalit job” as a man cleaning drains.

Experts say the bias comes from the training data these AI models learn from - huge swaths of unfiltered text and images from the internet, where caste prejudice still runs deep.

This is especially worrying because India is now OpenAI’s second-largest market, with millions of people using ChatGPT for writing, studying, and work. Without safeguards, subtle biases in the AI’s language could normalize discrimination in everyday use from job applications to classroom essays.

Millions in Space? Jeff Bezos Thinks So

Jeff Bezos says millions of people will live in space “in the next couple of decades.”

Speaking at Italian Tech Week in Turin, the Amazon and Blue Origin (a spacetech company) founder predicted a future where people live in massive space habitats, robots handle all the hard work, and floating AI data centers keep everything running.

He added that most people will choose space life “because they want to,” not because they have to.

It’s a bold claim, and it makes you wonder: Would you move to space if Bezos is right?