OpenAI Scientist Warns of Bioweapons and Grid Attacks as Company Ships GPT-6 Astra

OpenAI Scientist Warns of Bioweapons and Grid Attacks as Company Ships GPT-6 Astra

Omor Ibne Ehsan

Wed, September 9, 2026 at 9:30 PM GMT+3 5 min read

• NVDA -2.37%

Quick Read

• NVIDIA's Q2 revenue surged 106% to $96 billion, with Q3 guidance at $108 billion and 95 of 97 analysts remaining bullish on NVDA.

• Pachocki calls for voluntary AI slowdowns after flagging bioweapons and grid attacks as worst-case risks, yet no lab has reduced its release cadence.

• Just released. Our analysts combed the entire stock market and named the ten best stocks to buy right now, and NVIDIA made the cut. Enter your email to see the other nine names and why NVDA earned its spot. The report is free. Enter your email and see the full list.

The people building the most powerful AI systems on the planet are also warning that no one knows how to control them. That contradiction sits at the center of a CNBC segment aired September 8, 2026, in which reporter Kate Rooney walked viewers through an essay titled An Alien Mind written by OpenAI chief scientist Jakub Pachocki.

The essay lands the same week, per Rooney's CNBC segment, that OpenAI shipped GPT-6 Astra, and against a backdrop of 20 frontier-level AI models released in the past few months. CNBC also noted that Jensen Huang has said AGI has arrived.

For investors in NVIDIA (NASDAQ:NVDA), the picture is unusual. A senior scientist at OpenAI is calling for restraint, while OpenAI itself is committing to compute deployments that would make its NVIDIA bill one of the largest capital programs in corporate history. This is fundamentally a question about the infrastructure supplier.

The warning deserves to be priced into the risk framework around NVIDIA, even if it is unlikely to slow the buildout.

What Pachocki Actually Said

Rooney summarized the essay's core claim: "We don't fully understand how these models actually work. He talks about AI as something that is grown more than something that's designed."

Grown means the model's behavior emerges from training on data rather than being specified by an engineer. You get capability without a blueprint and discover what the system can do by testing it afterward.

Free Report, Just Released

Why NVDA Made Our Top 10 List

24/7 Wall St has helped investors make money for over two decades, and our top analysts just finished ranking the definitive Top 10 Stocks To Buy Now. Not the ten biggest companies. Not the ten everyone is arguing about. The ten best stocks to buy right now.

And NVDA made the cut. See who else did.

The report is free, and you can see why we think each stock is a top investment today.

Enter Your Email and See the Ten →

Rooney also relayed Pachocki's forward-looking concern: "These systems will see jumps in capability with consequences that no one is prepared for."

That framing matters because it comes from inside the lab whose next model people are worried about.

Alignment, Defined Properly

Alignment is the problem of getting a system to behave according to human values in situations it has never been shown during training. It gets harder as models get more capable.

Rooney was blunt about the state of the field: "No AI lab right now has solved this alignment issue."

Her summary of Pachocki's prescription was equally direct: "Voluntary slowdowns are going to be important until the industry agrees on safety standards."

Voluntary is the operative word. No lab has slowed down, and the release cadence Rooney described suggests none intend to.

Recursive Self-Improvement and the Named Harms

Recursive self-improvement is the scenario in which AI systems play a growing role in improving the next generation of AI systems. Each cycle compounds capability faster than human oversight can keep up.

Rooney enumerated the harms Pachocki flagged: "The worst case scenario is things like bioweapons, things on the level of national security that could be catastrophic. Cyber weapons, wide scale hacks, attacks on the grid."

These are tail risks rather than base cases, and they are why Pachocki argues for pauses even though his employer is shipping.

Is NVDA Stock a Buy?

Here is the commercial contradiction. NVIDIA's Q2 fiscal 2027 revenue was $96.22 billion, up 105.8% year over year, with data center revenue of $89.02 billion. Guidance for Q3 is $108.00 billion plus or minus 2%08.00 billion plus or minus 2%, disclosed in the company's 8-K filed August 26, 2026.

OpenAI's existing and planned commitments represent approximately 12 gigawatts of NVIDIA compute, and Jensen Huang told analysts NVIDIA now sees fiscal 2028 revenue growing approximately 70%, supply constrained, according to CNBC. Safety essays have not slowed that curve, and the order book shows no evidence they will. All of that compute also has to be powered, cooled, and networked by somebody, which is why we pulled together seven suppliers behind the buildout in a free AI infrastructure report.

Shares trade at $225.73, up 34.29% over the past year, with analyst sentiment showing 95 bullish ratings against 2 bearish. The alignment risk is real, and the current demand signal is louder.

A serious safety event at a frontier lab would likely hit NVIDIA's multiple before it hit revenue, which is the key risk to monitor alongside the demand backdrop.

Got $1,000? NVDA Made Our Top 10 List. Here's Who Else Did

If you have cash sitting in your account right now, give this two minutes. After more than two decades of helping investors beat the market, our top analysts at 24/7 Wall St. put together a definitive report on the Top 10 Stocks To Buy Today. And NVDA was one of them.

They combed the entire market. It's not 10 ideas, not 10 stocks everyone is talking about, it's what their research points to as the 10 best stocks to buy right now, and it's free. Read more here and see which stocks made the list -->>

Contact editorial@247wallst.com for any questions or corrections.