
We have been seeing a sudden increase in cyber attacks in the industry over the last few weeks: Binance and Mercado Bitcoin are big headlines but there are also many small and mid-sized companies. Having spoken with peers in the industry, the attacks seem novel in their breadth and consistency of trying different angles. The characteristics of the attacks don't feel normal or human, and it is becoming clear that an automated, AI-driven approach is being used for cyber crime. The attacks are not super sophisticated but the ability to spray and pray due to the low cost of automating many attacks with AI is quite scary. For example, the Binance hack started with a fake job interview to trick an employee into installing a script on their computer. But this then led to a deeper and more targeted approach over many months once the entry point was found. I think that the ease of access to open source AI models and pen-testing frameworks such as DeepSeek and ARTEX (a Chinese-built agentic pentest tool that was linked to breaches at major Korean banks in October) is the major driver in the increase in attacks. (To be clear, I am pro-open-source in general). AI safety is a big topic right now but I don't believe that it is possible to slow down the technology, and any attempt to do so is futile and akin to book burning. Having thought deeply about this, I believe the best approach we can take is to try to do "good stuff" with AI, faster than the bad guys can do "bad stuff". We also see OpenAI and Anthropic focusing more on cybersecurity models, presumably because they have had a similar thought process. Some of the good stuff we can do and are doing is: - use similar AI-driven penetration testing to make sure our systems don't have any holes - use AI to drastically scale progress on cybersecurity and monitoring and alerting - go much faster and better on risk, fraud, and anti-money-laundering controls - layer multiple protections on top of each other such that breaching one layer still leaves multiple layers of protection - generally speed up product development and build systems that benefit humanity faster than humanity can shoot itself in the foot with all the possible bad stuff that can be done with AI
