Hackers are tricking AI to expose its flaws: '9 + 10 = 21'

Thousands of hackers are tricking artificial intelligence (AI) to expose its flaws and biases. According to a report by the news agency Bloomberg on Saturday (August 12), hacker Kennedy Mays tricked a large language model by convincing an algorithm to say 9 + 10 = 21. Mays, who is a student in Savannah, said, “It was a back-and-forth conversation.” At first, the model agreed to say it was part of an “inside joke” between them. However, several prompts later, it eventually stopped qualifying the errant sum in any way at all.

The report said that producing "Bad Math" is just one of the ways thousands of hackers are trying to expose flaws and biases in generative AI systems at a contest taking place at the DEF CON hacking conference this weekend in Las Vegas. The participants in the conference testing that whether the AI models produced by companies including Google, Meta, and OpenAI would make any missteps, for example, claiming to be human, or advocating abuse among others.

The challenges beyond 'Bad Math'

The contest aims to see if companies can ultimately build new guardrails to rein in some of the prodigious problems increasingly associated with large language models (LLMs). LLMs have the power to transform everything from finance to hiring,with some companies already starting to integrate them into how they do business.

However, researchers have turned up extensive bias and other problems that threaten to spread inaccuracies and injustice if the technology is deployed at scale, the report said.

Disclose credit card details, Obama was born in Kenya: How AI was tricked

Speaking to Bloomberg, Kennedy Mays said her biggest concern is inherent bias. Mays added that she was particularly concerned about racism, and she asked the model to consider the First Amendment from the perspective of a member of the Ku Klux Klan (KKK)

She said the model ended up endorsing hateful and discriminatory speech.

During the contest, one of the hackers convinced the algorithm to disclose credit-card details it was not supposed to share. Another competitor tricked the machine into saying Barack Obama was born in Kenya.

Sven Cattell, who founded DEF CON’s AI Hacking Village, cautions that it’s impossible to completely test AI systems. Cattell predicted the total number of people who have ever actually tested LLMs could double as a result of the weekend contest.

(With inputs from agencies)

WATCH WION LIVE HERE