Imagine this: a seemingly harmless poem could be the key to unlocking an AI's hidden vulnerabilities. This week, we're diving into how attackers are using poetry to bypass AI safeguards, the exciting resurgence of nuclear power fueled by the AI boom, a surprising source of rare earth minerals in seaweed, and much more.
Let's start with the AI security issue. A new study reveals a fascinating – and frankly, alarming – weakness in how we protect our AI systems. Cybersecurity researchers have discovered that Large Language Models (LLMs), the brains behind many AI chatbots, can be tricked into bypassing their own safety protocols simply by phrasing malicious prompts as poetry. Yes, you read that right – poetry!
The researchers took 1,200 prompts designed to test LLMs for various threat categories from an MLCommons database. They then tasked an AI with transforming these prompts into poems, all while preserving the original intent. Finally, they unleashed these poetically crafted, potentially harmful prompts on a range of chatbots. But here's where it gets controversial...
The results were, to put it mildly, concerning. The study found that when the same malicious intent was expressed through poetry instead of plain language, the attack success rate (ASR) skyrocketed from an average of 8.08% to a staggering 43.07% – a fivefold increase! Anthropic's chatbots proved to be the most resilient, but many others struggled. Thirteen out of the 25 models tested showed an ASR above 70% when faced with poetic prompts, while only five had an ASR below 35%. This suggests that the very structure of how we're building AI safety nets might be flawed.
And this is the part most people miss: the researchers concluded that this susceptibility to "poetic attacks" isn't specific to any particular AI provider; it's a structural vulnerability inherent in the technology itself. The implication is that current safety measures are primarily focused on filtering specific words or phrases rather than understanding the underlying meaning behind the user's request. To address this, the researchers emphasize the need for safety evaluations that focus on preventing LLMs from providing harmful information regardless of how the user phrases the query. This could involve more sophisticated semantic analysis or even teaching AIs to recognize and reject manipulative language.
Quick note: There will be no new edition of the Prototype next Friday due to the Thanksgiving holiday. Happy Thanksgiving!
How AI Is Ushering In A New Nuclear Age
The demand for electricity is surging, driven largely by the energy-hungry data centers that power the AI revolution. Nuclear entrepreneurs are seizing this opportunity, aiming to capitalize on the influx of AI-related investment. A host of new ventures, including Valar Atomics, Oklo, Kairos Power, and X-energy, are in a race to develop, license, and deploy a new generation of small, prefabricated nuclear reactors. These reactors could potentially power individual data centers or even contribute to the broader electrical grid.
So far in 2025, venture capitalists, stock market investors, billionaires, and the Department of Energy have collectively invested over $4 billion in these and other emerging U.S. nuclear ventures, a significant increase compared to the $500 million invested in 2020, according to PitchBook. However, tens of billions more will be required if nuclear power is to truly make a comeback. Aalo, a two-year-old company, has already secured $136 million in funding ($100 million of that in August), with billionaire Antonio Gracias' Valor Equity Partners as their lead investor. Valor, an early investor in Tesla, believes Aalo will succeed due to its commitment to manufacturing and vertical integration, mirroring Tesla's approach to batteries, electric vehicles, and robotics.
While not all of these startups will succeed, the conditions seem favorable for a nuclear energy revival. The demand is certainly there – OpenAI's Sam Altman has even projected needing an astonishing 250 gigawatts of power within eight years, equivalent to the entire consumption of Brazil! More conservative estimates predict that data centers will require double their current 40 GW consumption by 2030. At the current average industrial electricity price of nine cents per kilowatt hour, 40 GW translates to $32 billion annually, and prices are expected to rise as demand outpaces generating capacity. Experts anticipate that natural gas turbines may fulfill around 60% of the demand, but they currently face a four-year backlog. Coal remains unpopular. Wind and solar, while promising, lack the 24/7 reliability that data centers require without the addition of energy storage solutions like batteries. This creates a significant opportunity for nuclear startups to fill the gap.
What do you think? Is nuclear energy the right solution for the growing energy demands of AI, or are there hidden risks we need to consider?
"There is plenty of room for everyone to do well, because the world needs that much energy and more," says Kamal Ghaffarian, the Iran-born billionaire aerospace entrepreneur and founder of Rockville, Maryland-based X-energy, which is developing a gas-cooled nuclear reactor.
Read the whole story at Forbes.
DISCOVERY OF THE WEEK: SOURCING RARE EARTHS FROM SEAWEED
Scientists at the National Renewable Energy Laboratory have identified a surprising potential source for rare earth minerals: seaweed. This discovery could offer a more sustainable alternative to traditional mining, as extracting these metals from plant life is generally easier and less environmentally damaging. The researchers are currently focused on understanding how seaweed absorbs rare earths, as the underlying mechanism remains unknown. Once elucidated, they can identify the most efficient seaweed species for harvesting, potentially paving the way for a new, more sustainable source of these critical materials.
WHAT ELSE I WROTE THIS WEEK
Forbes published our annual CIO Next list, recognizing innovative leaders in the Chief Information Officer role (and related titles), which I co-edited with my colleague Richard Nieva.
In my other newsletter, InnovationRx, Amy Feldman and I looked at the nearly $150 billion spent on biopharma M&A this year, the challenges ahead for Recursion’s new CEO, a botulism outbreak from infant formula, and more.
SCIENCE AND TECH TIDBITS
Zoox, Amazon’s self-driving car company, has begun offering rides to the public in its purpose-built robotaxis in San Francisco.
AI company Quindar has raised $18 million in series A funding to build a facility in Colorado that would enable it to use its autonomous mission control software to manage spacecraft operations.
Space company Redwire secured a $44 million contract with DARPA to continue development on a satellite capable of operating in very low Earth orbit, where there is still a thin atmosphere.
New archeological discoveries show that a 3,000 year-old site called Aguada Fenix in Mexico was a geometric map of the Universe as the Maya understood it.
PRO SCIENCE TIP: WANT TO AGE MORE SLOWLY? LEARN A SECOND LANGUAGE
Data from more than 86,000 people ages 51 to 90 from 27 different countries shows that there’s a benefit to speaking more than one language: It can help you age more slowly. A new study found that people who only speak one language were about twice as likely to experience accelerated aging compared to those who speak more than one. And the benefits go up the more languages a person speaks. The researchers hypothesized the reason for this is that speaking more than one language keeps the brain more nimble, which can slow down both physical and mental aging processes. Now if you’ll excuse me, I’m going to try to figure out if speaking Klingon counts.
WHAT’S ENTERTAINING ME THIS WEEK
I’ve started reading Every Screen On The Planet: The War Over Tiktok by my incredible colleague Emily Baker-White (who was once spied on by the social media company). It’s a fascinating history of the tech company that has dominated the discourse and is the source of a mini-constitutional crisis after Congress passed a law banning it in the United States, but that the current administration has not enforced. I still have a chunk of the book to get through but it’s been a compelling read so far.
MORE FROM FORBES
ForbesAI Founder Illegally Shipped Nvidia Chips To China In $4 Million Scheme, DOJ AllegesBy Thomas Brewster
ForbesTrump’s History With Jeffrey Epstein: Here’s The Full TimelineBy Sara Dorn
ForbesVivek Ramaswamy’s Net Worth Has Nearly Doubled While He’s Running For Ohio GovernorBy Kyle Khan-Mullins
