Altman says espionage against OpenAI has sometimes worked as a bugging case heads to trial Jan. 20
The OpenAI chief executive conceded espionage has sometimes worked and security has not been perfect, days after the bugging case got a January 20 trial date. Buyers of the $1.25 trillion private round, and holders behind Nvidia's $105 billion Ohio guarantee, are underwriting the perimeter that protects the model secrets themselves.
Vincent Jiangthree-minute read
Sam Altman has spent three years arguing OpenAI$1.18T — OpenAI, private, latest valuation $1.18T can be trusted with the most consequential technology of the age. On Semaphore's Future Proof podcast, the OpenAI chief executive admitted the opposite about his own walls: espionage against his company has sometimes worked. "I'm sure we've been tricked sometimes and I'm sure that our operational security and overall security has not been perfect," he said, naming the methods himself: a camera planted in a conference room, engineers taken out for drinks to extract pre-training secrets 1.
The admission lands days before a real test. On November 17, US District Judge Vince Chhabria set a January 20 trial for Abdulrahman Alfozan, the Saudi national and former OpenAI engineering lead accused of planting a recording device behind a television in an executive conference room 2. Prosecutors hold 160 hours of meeting recordings and remain locked out of Alfozan's phone 2. The device, per the FBI complaint, was a thumb-sized Seeed board with a detachable camera and microphone, retailing for $13.99 3.
Eleven billion in, on a wall its seller calls imperfect
OpenAI banked about $11 billion of new money in early November at a $1.25 trillion pre-money valuation, with up to $9 billion more of the round still unclosed 4. We reported the round on November 7, when the open question was whether the labs' balance sheets could support their compute commitments 4; what is new this week is the seller of that paper telling buyers, in his own words, that security has leaked 1.
Nvidia−3.58% — Nvidia, down 3.58 percent today, which guaranteed $105 billion for OpenAI's Ohio data center as the interview itself noted, is in turmoil of its own: Jensen Huang resigned November 12 after an affair with an alleged Chinese intelligence asset, and the board said this week it will not bring him back 5.
The money keeps coming anyway
Anthropic$350B — Anthropic, private, latest valuation $350B raised $81.3 billion in the largest IPO on record at $83 a share, a roughly $1.5 trillion fully diluted valuation 6.
Nvidia, one day earlier, reported $115.8 billion in revenue and $107.6 billion from data centers for the first quarter since Huang's exit, beating expectations 5.
Nvidia revenue rose every quarter through the leadership turmoil, to $115.8 billion
Data
| Revenue | |
|---|---|
| Q3 '24 | $35.08B |
| Q4 '24 | $39.33B |
| Q1 '25 | $44.06B |
| Q2 '25 | $46.74B |
| Q3 '25 | $57.01B |
| Q4 '25 | $68.13B |
| Q1 '26 | $81.62B |
| Q2 '26 | $96.22B |
| Q3 '26 | $115.8B |
So the capital keeps flowing, exactly as Altman predicted 1. The problem is what it is buying. If frontier model secrets are the asset behind a $1.25 trillion private valuation and a $1.5 trillion public one, the company selling the first just told buyers its own security has sometimes failed 1.
Altman calls his doctrine being "pragmatic centrists" on safety 1; an imperfect perimeter is the sharpest counterpoint to that pitch.
No one has shown that anything got out
There is a fair counterpoint the other way. Prosecutors are not aware of any evidence Alfozan engaged in espionage or stole anything, the prosecutor said in court: "At this point, I'm not suggesting the defendant stole anything" 2. OpenAI has said it found no indication he accessed or exfiltrated model weights, source code or customer data 3. The charge carries five years maximum 2.
The January trial, with FBI reports due to the defense by November 24 7, is the first scheduled reading on what, if anything, got out. Until then, every buyer of AI paper is pricing a perimeter its most-watched chief executive admits is not perfect.
Get Security by email, free
The breaches, flaws and security vendors AI is putting to the test. In our daily email, every weekday.






Reader comments
0 comments