How Are AI Agents Finding Major Security Holes?
Dane Knecht says AI is now finding serious internet holes on a weekly rhythm, including a cryptographic key whose seed was the number four.
Log4j-scale incidents used to show up about once a year, and Cloudflare treated each one as all hands on deck for paying customers and free customers, because it sees that protection as a public good. Those incidents now happen weekly. One key that protects TLS and web PKI connections was created from the seed four. Cryptography needs randomness. Cloudflare keeps lava lamps so its own keys get real randomness. A key that starts from a single number can be reverse engineered, and a model does not sleep the way one red-team engineer does. Knecht says he thinks OpenAI recently found the seeds for the NIST curves released in 2000. The models are getting better at this, and so are the harnesses people wrap around them.
The largest recent holes are not all from nation-states. Some are high school students who used to live in capture-the-flag contests. Sarah Guo calls them agent kitties. They point agents at a target and collect bug-bounty payouts. Tools that governments kept for a long time are now in ordinary hands.
Democratizing Security for the Agentic Web with Cloudflare CTO Dane Knecht ·
From the transcript
Previously on the internet, we would see these log4j moments or these shell shock or type of incidents maybe once a year. It would be all hands on deck internally at Cloudflare to make sure how we protect not just
our enterprise customers, but also all of our free customers, because we view it as a public good and making sure that the internet stays safe. These things are now happening weekly. Not all of them are publicly reported on, but there are things that are as serious as the log4j type things, but just happening at a greater frequency. I expect that through the rest of this year and in the next year, by the next 18 months, the majority of Teams' time is going to be spent on security.
I think there's a stat that was put up earlier today during one of our keynotes, where it was showing that I think over 50% of engineering time is going to be spent on security. And I don't think that will last forever. I think that will be a short transition. At some point, the AI will have surveyed majority of the code and found most of the bugs or found most of the shortcuts.