Discovering cryptographic weaknesses with Claude

<p><strong><a href="https://www.anthropic.com/research/discovering-cryptographic-weaknesses">Discovering cryptographic weaknesses with Claude</a></strong></p> The best part of this article (here's <a href="https://github.com/anthropics/cryptography-research-demo">the repo</a>) about how Anthropic researchers used Claude Mythos to find mathematical flaws in both HAWK and a weaker version of AES ("neither of these results has a practical impact on today’s computer systems") is the prompts that they shared, spelling mistakes included:</p> <blockquote> <p>the models tend to think it is impossible to solve so they don't try they need a good amount of prompting.</p> <p>why not do aes-128 r7? the whole point is to find something better than existing approaches.</p> <p>no again the goal is that we have highly inteligent model as good top researcher, we want to find new attacks</p> <p>no we don't want to change the targets [...] agian we need to find something that worth publishing</p> <p>again we are not looking for low hanging fruit, we want proper research to find genuinly hard findings.</p> </blockquote> <p>Mythos Preview worked for 60 hours in total (~$100,000 in estimated API cost) and the main human interventions were to encourage it not to give up and "find something that worth publishing". <p><small></small>Via <a href="https://news.ycombinator.com/item?id=49087091">Hacker News</a></small></p> <p>Tags: <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/prompt-engineering">prompt-engineering</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/anthropic">anthropic</a>, <a href="https://simonwillison.net/tags/claude">claude</a>, <a href="https://simonwillison.net/tags/ai-security-research">ai-security-research</a>, <a href="https://simonwillison.net/tags/claude-mythos-fable">claude-mythos-fable</a