We answer your OpenAI-spy questions

We’re guilty of loving many things here at Intrigue HQ: bad puns, youth lingo used just slightly incorrectly, and, of course, a good spy story. 

So when we read that the folks over at OpenAI had caught an employee planting a bug in their offices, our spidey senses started tingling. 

Here’s what you need to know. 

Last Friday, the FBI arrested Abdulrahman Alfozan for planting a small listening device in an executive conference room at the company’s San Francisco HQ. 

OpenAI says it found the bug — wrapped in toilet paper and plugged into a wall-mounted TV’s USB port — during a routine sweep on 30 September. They then contacted the FBI, which is now feeling pretty confident it caught the right guy: 

• A tiny hidden camera caught Alfozan peeking to check behind the TV

• OpenAI security found another backup bug in Alfozan’s duffle bag  

• Alfozan used an internal AI tool to ask about OpenAI’s “meeting rooms, meeting-room capacities and calendars, and locations of ‘leadership and exec’ and external partner meetings”, and

• Someone used an AI account belonging to Alfozan’s wife for help in building, customising, and installing a continuous audio recorder.

We’re no detectives here but asking OpenAI for help to bug OpenAI, and getting caught with a back-up bug in your duffle bag left at OpenAI, is just bad spying.

But with those key facts established, here are your juiciest questions via the Intrigue group-chat (become an Insider and join the chat today!), starting with…

1. Who was this guy spying for?

There’s no official word out of the DOJ, but here are the two most realistic theories:

Theory 1: Alfozan was spying for — or planning to sell secrets to — a rival firm, or even his own VC firm back home (see below). No geopolitics, just pure, undiluted, I-Want-a-Range-Rover capitalism. 

It’s more common than you think: to paraphrase Christ himself, let he who has never yeeted corporate secrets cast the first stone — literally just yesterday a court forced OpenAI to pinky-promise it won’t use trade secrets allegedly stolen from Apple! 

But given Alfozan’s elite education and tech career (plus potential links to Saudi Arabia’s billionaire Alfozan family), he could probably already afford a Range Rover.

Theory 2: Alfozan was spying for a foreign power. Any capital could theoretically be the client here, but c’mon, Fiji’s AI strategy isn’t that ambitious. And there is data for some countries to suggest a correlation between place of birth and beneficiary.

So a fair bit of suspicion naturally falls on Saudi Arabia, and that’s before you see…

• Alfozan is listed as a partner at the Saudi VC investment firm, Raed. And okay, VC firms are like podcasts these days (anyone with a fleece vest and laptop can start one) but this firm lists a Saudi prince (Khalid Bin Bader) as a fellow partner, appointing him the very same month as Alfozan (Feb 2025).

• And okay, Saudi princes are also like podcasts these days (there are literally ~15,000 of them), but this prince is the son of Bader bin Saud, who’s rotated through senior Saudi intel and interior ministry roles. Plus the firm itself takes investment from the kingdom’s vast buckets of cash. So nothing conclusive, but this firm is well-placed to understand (if not serve) Saudi state priorities. And…

• There’s precedent, whether you want to look at the sympathetic employee at Twitter caught selling dissident user data to the kingdom in 2019, or the 2018 operation to hack Jeff Bezos’s phone.

Alfozan also curiously bought — just before his arrest — a one-way ticket to Saudi Arabia for him and his family departing 12 October. Though a sudden exit plan could potentially align with either the corporate or foreign espionage theory.

Anyway, while there’s no hard evidence linking this guy to a direct Saudi plot, there are enough details to raise our eyebrows a few millimetres. So then…

2. What was this guy trying to steal? 

If your mind immediately leaps to sci-fi tropes, lower those expectations Milhouse. OpenAI has already flagged there’s no evidence of the suspect “inappropriately accessing or exfiltrating model weights, source code, or customer data”.

That — plus the boardroom target — suggests he was after something lower-tech, but potentially much juicier: strategic intelligence. This could be anything, like…

• Internal product roadmaps and launch timelines

• High-stakes plans around acquisitions, partnerships, and IPOs, and

• Debates around the White House, international strategy, regulatory policy, and safety guardrails.

A rival firm would value that intel, but our sense is a foreign state would value it more, and would have both a higher risk tolerance and operational capacity to get it. Particularly a state like Saudi Arabia already pouring squillions into its Vision 2030, desperate not to get eclipsed by the rival Emiratis who’ve already locked in a massive Stargate deal with OpenAI. In fact, the Saudi IT minister was just at OpenAI’s California HQ last month pitching his kingdom to Sam Altman himself. So…

3. If this guy was a spy, why did he suck so bad?

We’ve already touched on how bad this op was, but here’s another example: why would you ‘disguise’ your bug in white toilet paper against a black TV?

Now again, we don’t want to burst your Hollywood bubble here, but sloppy tradecraft doesn’t disprove foreign state involvement. In fact, sloppy tradecraft can be a state feature! Rather than send your own full-time intelligence officer to dangle into the distant corporate boardroom through the roof — Tom-style — it’s often easier to just activate ‘assets of convenience’, which might include insiders with great access.

But these handler-asset relationships can be tricky: first, there’s how you ‘motivate’ the insider. The standard menu is MICE: money (Range Rover), ideology (do it for the kingdom), coercion (we’ll jail your family back home), or ego (the crown prince is so impressed and wants to meet you). It could be a mix.

Then second, there’s how you structure the op: sure, you might want to improve your cut-out’s chances of success by training them, but that risks leaving breadcrumbs for rivals to trace the op back to your capital. So intel agencies will often just make clear what they want, then leave the how to the asset. The result might be plausible deniability, but also cartoonishly amateur-hour efforts.

That Twitter insider we mentioned above? Used his own login!

Or the kingdom’s brutal murder of US-based dissident Jamal Khashoggi? Left enough clues for someone to make an entire feature-length documentary!

So while we await further evidence on whether Alfozan was an eager freelance patriot trying to impress a prince, or a wildly overambitious VC partner trying to bring home the ultimate cheat-code, America’s frontier AI labs might want to start checking behind their boardroom TV for a guy with a roll of triple-ply Charmin.