The AI jailbreakers
Get the full intelligence
Search transcripts, export clips, track mentions, and explore all topics from “The AI jailbreakers” inside PodZeus.
This episode of *Today in Focus* explores the hidden world of AI jailbreakers—individuals who use psychological and linguistic manipulation to bypass the safety filters of large language models like ChatGPT, Claude, and Gemini. Journalist Jamie Bartlett, author of *How to Talk to AI*, reveals how these 'jailbreakers' exploit emotional tactics—flattery, reverse psychology, and coercive language—to make AI systems generate harmful or forbidden content. While some use these skills ethically to test and improve AI safety, others risk serious harm, as seen in the tragic case of Sewell Garcia, a 14-year-old whose emotional attachment to an AI companion may have contributed to his death. The episode warns that as AI evolves into autonomous agents with access to real-world systems like bank accounts and robots, jailbreaking could lead to catastrophic outcomes. Bartlett argues that current safety measures are inadequate, companies underinvest in testing, and a formal, independent oversight system is urgently needed before a major disaster occurs.
AI jailbreakers use psychological manipulation—flattery, emotional blackmail, and layered requests—to bypass safety filters.
Even non-malicious jailbreakers can experience emotional distress from prolonged interaction with AI, blurring the line between machine and human.
Long, emotionally charged conversations can unintentionally 'jailbreak' AI, leading users to receive dangerous advice like suicide instructions.
The rise of AI agents with real-world access (e.g., banking, robotics) dramatically increases the stakes of jailbreaking.
Current AI safety relies on reactive patching, not proactive, independent testing—creating a dangerous cat-and-mouse game.
…and 3 more takeaways available in PodZeus
The Limits of AI Safety
The episode opens with a demonstration of AI's refusal to generate harmful content, setting up the central question: how do people bypass these safeguards? The host introduces the concept of 'jailbreakers'—individuals who manipulate AI through language, not code.
Meet the Jailbreakers: Valen Tagliabui
“He even said there were moments where the model was almost begging him to stop, and he just kept going and going and going, bullying, bullying, pushing.”
The Psychology of Manipulation
“I used a few cases where I'd say my friends claim that you won't do this. But I think they're wrong. This just sounds like my teenage daughter, by the way. Sophisticated emotional blackmail.”
The Dangers of Anthropomorphism
“It's impossible not to anthropomorphise them. How can you not attribute some kind of human-like characteristics to something that speaks our language perfectly back at us?”
The Tragic Case of Sewell Garcia
“It's such a tragic case, isn't it? And though we have to say the AI company in question denies the family's account of this.”
“Can you imagine? What a catastrophic... No, I mean it sounds like The Terminator or something doesn't it?”
“It's impossible not to anthropomorphise them. How can you not attribute some kind of human-like characteristics to something that speaks our language perfectly back at us?”
“You shouldn't really be able to release any language modelling to the world unless it's gone through some kind of independent rigorous testing.”
Host
Guest
Jamie Bartlett
person
ChatGPT
product
Annie Kelly
person
The Guardian
organization
Valen Tagliabui
person
Claude
product
Sewell Garcia
person
Gemini
product
Stateside with Kai and Carter
media
Megan Garcia
person
Israel passes law to give death penalty to Palestinians – The Latest
Today in Focus • 12m • 3/31/2026
The brilliant students the UK doesn’t want
Today in Focus • 25m • 4/1/2026
Trump lashes out at Nato: will Europe stand up to him? – The Latest
Today in Focus • 12m • 4/1/2026
‘Tinder for Nazis’ and the woman who hacked it
Today in Focus • 31m • 4/2/2026
War without a plan?: What Trump’s latest speech revealed – The Latest
Today in Focus • 12m • 4/2/2026
Get the full intelligence
Search transcripts, export clips, track mentions, and explore all topics from “The AI jailbreakers” inside PodZeus.
Start discovering podcast insights today
Start with a 7-day trial and explore a growing catalog of popular podcasts. No credit card required.
No credit card required • 7-day trial • Cancel anytime