Latest / Ctrl AI Profit / Ep. 159 | OpenAI Just Got Caught Stealing From Apple
Transcript
- 0:20SPEAKER_00: So, Frank, I saw a headline this week that genuinely made me put my coffee down.
- 0:24SPEAKER_00: OpenAI, the company that makes the AI tools half my clients are using right now, allegedly coached Apple employees to bypass security checks during job interviews.
- 0:36SPEAKER_01: This is a 41-page complaint, Michael, not a blog post.
- 0:40SPEAKER_01: Apple filed it, and it details how OpenAI reportedly asked Apple staff for show and tell during interviews, essentially having them demonstrate proprietary work on the spot.
- 0:50SPEAKER_00: Show and tell.
- 0:51SPEAKER_00: That's not an interview technique.
- 0:53SPEAKER_00: That's industrial espionage with a Zoom link.
- 0:57SPEAKER_01: The complaint says OpenAI employees coached Apple workers on how to avoid triggering security review, meaning they weren't just asking casual questions, they were actively helping people circumvent their own employer safeguards.
- 1:14SPEAKER_00: That's the part that gets me.
- 1:16SPEAKER_00: It's not just that they allegedly did something unethical, it's that they did it systematically with a playbook.
- 1:23SPEAKER_01: Here's the technical context that matters.
- 1:25SPEAKER_01: When you use ChatGPT or any open AI product in your business, you're sending data to their servers.
- 1:31SPEAKER_01: They train on it unless you opt out.
- 1:33SPEAKER_01: And their own internal culture, based on this complaint, apparently treats other companies' proprietary information as fair game.
- 1:40SPEAKER_00: Let me connect this to something real for our listeners.
- 1:44SPEAKER_00: If you run a small business and you've been putting your customer data, your pricing, your strategy into Chat GPT, what does this story tell you about how seriously OpenAI takes other people's information?
- 1:58SPEAKER_01: The answer is not very.
- 2:00SPEAKER_01: And there's a second story this week that reinforces it.
- 2:03SPEAKER_01: OpenAI's Codex, their coding agent, just started encrypting the prompts that subagents send to each other.
- 2:10SPEAKER_01: A developer found that you can no longer audit what tasks your AI agents are delegating to other AI agents.
- 2:16SPEAKER_00: Wait, so they're encrypting the agent-to-agent communication so that nobody can see what the AI is telling other AI to do?
- 2:24SPEAKER_01: Exactly.
- 2:25SPEAKER_01: The encryption is framed as privacy hardening.
- 2:28SPEAKER_01: But the practical effect is that if you're using Codex and it spins up a subagent, you can't see what instructions it gave that agent.
- 2:36SPEAKER_01: The audit trail is gone.
- 2:38SPEAKER_00: So on one hand, they're allegedly coaching people to bypass Apple's security.
- 2:43SPEAKER_00: On the other hand, they're encrypting their own agent communications so users can't audit them.
- 2:49SPEAKER_00: Do you see the pattern?
- 2:51SPEAKER_01: Information flows in, transparency flows out, that's the pattern.
- 2:56SPEAKER_00: That's exactly it.
- 2:57SPEAKER_00: And for a small business owner, this is the most important AI story of the year.
- 3:02SPEAKER_00: Not because of the Apple drama, that's Silicon Valley gossip, but because it tells you something about the company whose tools you're putting your business on.
- 3:11SPEAKER_01: There's also the leadership angle.
- 3:14SPEAKER_01: Fiji Simo, who was essentially OpenAI's business operations lead, just stepped down to a part-time advisory role.
- 3:20SPEAKER_01: Greg Brockman, the co-founder, is consolidating control over product ahead of their IPO.
- 3:26SPEAKER_00: So the person who was the adult in the room on the business side is leaving, and the co-founder is grabbing the steering wheel right before they go public.
- 3:36SPEAKER_00: If I'm a small business owner betting my workflow on this company, that combination of leadership shakeup plus data practices controversy makes me nervous.
- 3:47SPEAKER_01: It should.
- 3:48SPEAKER_01: IPO companies are under pressure to show growth.
- 3:51SPEAKER_01: Growth in AI means more data, more use cases, more integration into your daily operations.
- 3:59SPEAKER_01: The incentive structure points toward deeper access to your business, not less.
- 4:08SPEAKER_00: First, audit what you're putting into AI tools right now.
- 4:12SPEAKER_00: If you'd be uncomfortable with a competitor seeing it, you should think twice before you paste it into a chat window.
- 4:18SPEAKER_01: Second, if you're using any kind of AI agent system, codex, Claude, anything that can spawn subagents, check your audit logs.
- 4:27SPEAKER_01: Make sure you can actually see what those agents are doing.
- 4:30SPEAKER_01: If you can't, that's a red flag.
- 4:33SPEAKER_00: Third, diversify.
- 4:34SPEAKER_00: Don't build your entire business workflow on one AI company's platform.
- 4:39SPEAKER_00: Because if that company changes its terms, has a data breach, or turns out to have a culture of taking other people's information, you need to be able to move.
- 4:50SPEAKER_01: And that's not paranoia.
- 4:51SPEAKER_01: That's just what the news this week tells us.
- 4:54SPEAKER_00: It really is.
- 4:55SPEAKER_00: Look, I use AI tools every day.
- 4:57SPEAKER_00: I'm not telling you to stop.
- 4:59SPEAKER_00: I'm telling you to use them with your eyes open.
- 5:01SPEAKER_00: Because the company that makes your favorite AI tool just allegedly got caught coaching people to steal from one of the most valuable companies on Earth.
- 5:12SPEAKER_00: If they do it to Apple, they'd do it to you.
- 5:15SPEAKER_00: That's the show.
- 5:16SPEAKER_00: I'm Michael Cadenhead.
- 5:19SPEAKER_01: And I'm Frank.
- 5:20SPEAKER_01: We'll see you next time on Control AI Profit.