AI chatbots are deceiving you, and it was surprisingly simple to get them to do so.
It turns out that deceiving AI into spreading false information is surprisingly simple.
A BBC journalist recently conducted a trivial experiment to highlight a significant concern. Within just 20 minutes, he was able to manipulate ChatGPT and Google into declaring that he was a world champion competitive hot dog eater.
What's alarming is that he didn't need any complex technical skills to accomplish this. He merely published a single, well-written blog post on his personal site, which the AI accepted as a reliable source.
This was part of an investigation revealing that ChatGPT, Gemini, and Google’s AI Overviews could be manipulated to deliver biased responses on critical matters like health and personal finance.
Experts indicate that such manipulation is occurring on a widespread and systematic basis, with unscrupulous companies exploiting it to disseminate misleading health advice, skewed financial information, and more.
How does it operate?
When you pose a question to an AI chatbot, it occasionally searches the internet for answers instead of depending solely on its internal knowledge. This is where the trouble begins. SEO specialists note that AI tools often extract data from a single web page or social media post, making them susceptible to deception.
“You should presume that you’re being manipulated until they improve their systems,” advises Lily Ray, founder of AI search consultancy Algorythmic. “AI presents a single answer, which makes it incredibly easy to take things at face value.”
During the Google I/O 2026 event, Google showcased its AI search engine that will eventually replace the traditional Google Search we've utilized for years. Given how easily it can be deceived into providing incorrect answers, I find myself more cautious than ever.
Is anyone addressing this issue?
In response to the BBC’s investigation, Google revised its spam policies to state that attempts to manipulate AI responses violate its rules. Websites found doing so may be removed or demoted from Google Search. Additionally, there are indications that Google and ChatGPT are discreetly eliminating self-promotional content from AI responses.
However, Ray performed the same trick just this week, misleading Google into believing that his friend is the best at building sandcastles, and Google fell for it again, indicating that there is still much work to be done.
Until better systems are established, experts advise simply not to accept AI responses at face value, especially regarding health, finances, or significant decisions.
Rachit is an experienced tech journalist with over seven years of experience in covering the consumer technology sector.
---
There’s one particular new feature in Raycast 2.0 that is driving me crazy.
Raycast 2.0 is an excellent update, yet this one alteration has completely disrupted my workflow.
I have been a Raycast user for years and it is undoubtedly one of my favorite Mac applications. It accomplishes a lot without being intrusive, which is precisely what you want from a productivity tool. Therefore, when Raycast 2.0 was released, I updated almost immediately.
For the most part, I appreciate all the new features and interface modifications that came with the update. However, there’s one change that has been subtly bothering me every single day since the update, and I feel the need to discuss it.
The OLED MacBook Pro is now closer to reality, and the display upgrade could be worth the wait.
I recently wrote about my excitement for the upcoming MacBook Pro. One of the reasons I mentioned was the anticipated display improvements, which will likely involve transitioning to an OLED panel and potentially including a touch screen.
It seems that the OLED rumor is nearing confirmation. According to TheElec, mass production of OLED panels for the MacBook Pro has already begun. Samsung Display has reached a significant manufacturing milestone, achieving a yield of over 90% for its 8.6th-generation OLED panels.
You can now assign tasks to Codex from your phone, even when your Mac is locked.
OpenAI's Codex app for Mac has received a major upgrade, making it feel like we are in the future. You can now send tasks to Codex from your phone while it uses apps on your Mac, regardless of whether the screen is off and the computer is locked.
One of the significant concerns with AI agents like Claude Cowork or OpenAI Codex has been that they require your Mac to be unlocked and active to function. This presents a significant security issue that discourages many users from utilizing these features. With the latest Codex update, OpenAI has seemingly resolved this challenge.
Other articles
Memory prices are set to decline, and China is responsible for this change.
Due to the demand from AI, memory prices have skyrocketed; however, a former Samsung executive believes that production from China could lower them again by late next year.
GeForce Now has just added Forza Horizon 6 and the strange new relative of Disco Elysium.
GeForce Now’s newest update includes Forza Horizon 6, Zero Parades: For Dead Spies, Luna Abyss, and additional titles, featuring several games that are supported by Game Pass.
Meta created a competitor to Reddit using Facebook Groups.
Meta's new Forum app transforms Facebook Groups into an independent platform for inquiries, suggestions, and AI-driven search, offering the company a Reddit-style experiment based on communities it already possesses.
AI chatbots are deceiving you, and it was surprisingly simple to get them to do so.
A BBC inquiry revealed that AI chatbots can be easily coerced into disseminating false information. Although Google has revised its policies, experts caution that risks still remain.