Bloomberg Talks · · 18 min

OpenAI Foundation Co-Founder & President Greg Brockman and Cloudflare CEO Matthew Prince Talk AI & Cybersecurity

With Greg Brockman, Matthew Prince

Summary

Bloomberg Talks hosts OpenAI co-founder and president Greg Brockman and Cloudflare CEO Matthew Prince on how AI is reshaping cybersecurity. The central example is a PHP flaw running on roughly 70% of websites that had existed for years; OpenAI's tools found it and it was disclosed responsibly to Cloudflare and others, who protected customers before public disclosure. Prince argues AI only uncovered an existing bug, leaving software more secure, and expects defenders with access to tools from OpenAI and Anthropic to outpace attackers, though he warns the flaw is now public and Cloudflare sees millions of exploit attempts per minute, so PHP users must patch immediately. Brockman describes a 'defender's window' in which frontier models run ahead of broadly available tools, and cites a Calif exploit where OpenAI models plus a researcher's expertise did most of the work, followed by an internal model autonomously finding a more complete version. Both acknowledge that patching at scale is becoming harder, that the PHP fix required about two months of manual coordination, and that the system needs to change, though Brockman admits they do not yet have the answers.

Key takeaways

  • Matthew Prince on Cybersecurity: AI found a years-old latent PHP flaw; it did not create it, and fixing it makes the internet more secure.
  • Greg Brockman on Cybersecurity: Attackers will ramp up AI-driven exploitation of these flaws; attempts so far are not yet at massive scale.
  • Greg Brockman on Cybersecurity: Growing AI-found vulnerability volumes will need a new societal patching process; the answers are not yet known.

What was said, by company and theme

3 views from 2 speakers. Every quote is checked against the episode's transcript (how).

Cybersecurity

  • Matthew Prince · Cloudflare

    AI found a years-old latent PHP flaw; it did not create it, and fixing it makes the internet more secure.

    Prince argues AI is the real hero. The PHP flaw, which runs on roughly 70% of the world's websites, had existed for years and may have been exploited by nation states or hackers. OpenAI's tools found the existing bug and disclosed it responsibly to Cloudflare and others, who then protected customers before public disclosure. He says AI did not create the bug, only found it, so the software is now significantly more secure. He is optimistic that defenders with more resources and access to tools from OpenAI and Anthropic will find and fix more flaws than attackers.

    “AI is the real hero of this story. These flaws have existed for quite some time. They're extremely serious.”at 5:22
  • Greg Brockman · OpenAI

    Attackers will ramp up AI-driven exploitation of these flaws; attempts so far are not yet at massive scale.

    Brockman says OpenAI believes attackers will broadly use models to exploit these flaws. OpenAI is monitoring for it and has seen attempts in the past, though not at massive scale yet. He expects a ramp-up and points to Cloudflare reporting a very steep increase in attempted exploitation. He says the defense is responsible disclosure and repairing the internet before attackers act.

    “We believe that attackers are going to start using models broadly to try to exploit these flaws... I think we're starting to see attempts or we have seen attempts in the past, but not at massive scale yet.”at 6:43
  • Greg Brockman · OpenAI

    Growing AI-found vulnerability volumes will need a new societal patching process; the answers are not yet known.

    Brockman says that as AI finds more vulnerabilities, it gets harder for anyone to apply every patch. The PHP response took about two months of manual work with partners and government. He cites precedent: the FBI obtaining court orders to clear botnet malware, because an insecure internet affects everyone, not only the site owner. He says the system needs to change, but he does not think they have the answers yet, and the question is how to keep defenders ahead through some societal process.

    “And so I think that the core question we have, I don't think that we have the answers yet”at 11:47

Topics

  • AI-discovered software vulnerabilities
  • AI-enabled cyberattacks
  • PHP flaw exploitation
  • autonomous vulnerability discovery
  • defender's window
  • vulnerability patching at scale

More from Bloomberg Talks

All Bloomberg Talks episodes

Ask your AI about this episode

Chorus gives Claude, ChatGPT and other assistants every view from this episode and the rest of the market's podcasts. Ask who agrees, who changed their mind, and what else was said about the same companies.

Setup guide

Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus