Tools

Search

Explore

Videos Channels Figures

Atmrix

About

Tools

Search

Explore

Videos Channels Figures

Atmrix

About

SS

Sander Schulhoff

06/19/25

@ Lenny's Podcast

Common defenses against prompt injection, like improving prompts or using guardrails, often do not work effectively against motivated attackers.

Video

LP

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff

@ Lenny's Podcast

AI prompt engineering in 2025: What works and what doesn’t | Sander Schulhoff

06/19/25

Related Takeaways

SS

Sander Schulhoff

06/19/25

@ Lenny's Podcast

Prompt injection involves manipulating AI to produce harmful outputs, which remains a significant challenge in AI security.

SS

Sander Schulhoff

06/19/25

@ Lenny's Podcast

Safety tuning and fine-tuning models on specific tasks can help mitigate the risks of prompt injection by making models less susceptible to harmful prompts.

SS

Sander Schulhoff

06/19/25

@ Lenny's Podcast

Prompt injection techniques can trick models into generating harmful content, and these techniques continue to evolve and pose security risks.

BJ

Bihan Jiang

02/20/25

@ LangChain

We've built in security measures to prevent prompt injections and ensure that AI agents behave as expected, especially in regulated environments.

YC

Y Combinator Cast

12/07/24

@ Y Combinator

Claude has built-in guardrails to prevent misuse, avoiding actions like account creation or social media content generation, but it remains vulnerable to prompt injection attacks.

SS

Sander Schulhoff

06/19/25

@ Lenny's Podcast

Prompt engineering is crucial; bad prompts can lead to 0% accuracy, while good prompts can boost performance to 90%.

JB

Joe Benton

03/18/25

@ Anthropic

A misaligned AI model might pretend to be incompetent during control evaluations to avoid being deployed with better mitigations.

AA

Amanda Askell

09/06/24

@ Anthropic

It's crucial to think about edge cases when designing prompts, ensuring they work not just for typical scenarios but also for unusual inputs.

JB

Joe Benton

03/18/25

@ Anthropic

Mitigations suggested for AI control often overlap with those that responsible AI developers already implement to ensure robustness.