Ciberseguridad: cuidado con lo que le dices a la IA

The talk provides a clear and accessible overview of cybersecurity risks associated with AI chatbots, making it valuable for a general audience. The speaker effectively uses real-world examples to illustrate abstract concepts, such as the Samsung incident and the Dutch researcher’s discovery of exposed conversations. These examples are well-known and add credibility to the claims. However, the talk lacks depth in technical explanations; for instance, it does not delve into how LLMs actually process data or the specifics of prompt injection attacks. The speaker also does not provide citations or references to academic sources, which limits the scientific rigor. The interactive activities, such as the prompt injection game, are a strong pedagogical tool, but their outcomes are not described in detail. The talk’s strength lies in its practical advice, such as anonymizing data and verifying AI outputs. The title accurately reflects the content, and the talk stays focused on its topic. Overall, the information is reliable and well-presented, but it would benefit from more technical detail and source citations for a more rigorous scientific analysis.

The title accurately reflects the content: a cybersecurity talk focused on risks of sharing information with AI.

7/10

The talk is based on well-known cybersecurity risks and real incidents (Samsung 2023, Dutch researcher 2025) but lacks detailed citations or technical depth. The speaker is an expert (likely from a university) and the content is accurate but presented in a simplified manner.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction to the talk and the topic of AI cybersecurity.
• 02:00Explanation of what LLMs are and how they work.
• 05:00Discussion on the widespread use of AI in daily life.
• 10:00First risk: privacy and data leakage.
• 15:00Real case: Samsung employees leaking data via ChatGPT.
• 20:00Second risk: hallucinations and false information.
• 25:00Third risk: phishing and AI-generated scams.
• 30:00Fourth risk: prompt injection and how to protect against it.
• 35:00Interactive activity: prompt injection game.
• 40:00Best practices for safe AI usage and conclusion.

• UBUInvestiga blog — Official blog of the university research outreach channel, mentioned in the video description.
• UBUInvestiga YouTube channel — Channel hosting the talk, mentioned in the description.

• OWASP Top 10 for LLM Applications — Lists prompt injection and sensitive information disclosure as top risks, aligning with the talk.
• Samsung ChatGPT leak news — BBC article covering the Samsung incident, confirming the event.

• OpenAI's data usage policy — OpenAI states that they do not train on API data by default, but the talk suggests otherwise for free users. This discrepancy may be due to changes in policy over time.

• www.youtube.com