Substack has launched a new feature that can show you which of your favorite newsletters are being written using AI.

This week, the newsletter and writing platform announced an integration with the AI writing detection software Pangram that will allow users to scan posts, comments, and replies on Substack’s app to see an estimate of how much of the content was written by a human and how much was AI.

In the short term, the move might be bad for Substack’s business, as it could expose many of the newsletters on its platform that aren’t entirely written by people. That could potentially erode trust in the platform’s ecosystem of independent news and blogs, or even damage its reputation as a host of high-quality content.

But in the long term, AI detection features could help keep Substack free of “AI slop,” and encourage more users to trust what they’re reading was written by a person, or at least better understand when it’s not.

Substack joins several platforms that are leaning towards labeling AI content as such, especially now that AI is playing a greater role in the creation process. Photos and videos generated with AI are labeled as such on social media sites, while music streaming services have more recently begun labeling and, in some cases, penalizing AI-generated music.

“This is good use of AI,” Substack CEO Chris Best said.

“When I used to pitch Substack to writers, one way I would do it is […] we’ll do everything for you except the hard part,” he explained in an online chat with Pangram’s founder, Max Spero. “You have to have something — an idea that’s worth reading, that’s worth caring about, that’s worth sharing. That one thing is very hard and very valuable […] software should do everything else, but I think you do want the person to do the hard part.”

The feature will be available in Substack’s app for any post, note, reply, or comment above 100 characters. Substack will also allow its writers to include an optional AI Author’s note, using which creators can properly disclose their use of AI, the company told TechCrunch.

The company clarified that the tool is not meant to prohibit or penalize AI-assisted writing, but rather to encourage writers to add a “how I make this” statement, where they explain their process.

Publishers can also run Pangram on their own drafts before publication, and report and remove scans on their own work they believe are mistakes.

As Chinese open-weight AI models grow in capability and popularity, arguments about what should be done about them have once again reached a fever pitch.

There’s talk that the Trump administration might try to ban them (though it hasn’t yet acted on the idea). Meanwhile, proprietary model makers, particularly OpenAI and Anthropic, appear increasingly concerned about them.

Open-weight models such as Moonshot AI’s Kimi K3 or Alibaba’s Qwen offer inference at a fraction of the token cost of closed-source models from these large U.S. labs. The fear is that they also pose some sort of threat. Certainly, they threaten the profit margins of the large proprietary AI labs.

But should enterprises running these models in their own data centers succumb to the fear that they could be a vector for Chinese hackers?

No, says Lucas Atkins, the CTO of Arcee, which is building open models to give U.S. companies a homegrown alternative to Chinese models.

If any startup would benefit from a ban on Chinese models, Arcee would. But Atkins says China’s open models are no more dangerous than any other open-source software a company may use. In fact, he says, they even offer benefits even to his own company.

“A lot of people view this as similar to a Chinese software program. Like, it was coded with these x, y, z intentions” that a bad actor could simply command, he said.

“That is fundamentally not how these models are trained. There is really not any way for an Arcee, or an Alibaba, to make a model, have someone run it in their own environment and for us have any access to it whatsoever,” he explained.

While most of these models are what’s known as “open weight,” and are not really fully open-source software, the source code (the part that will actually run on servers), if it is downloaded from open-source sites like Hugging Face, is similarly largely visible and reviewable. (What isn’t available is the methods and data used to train the models.)

Large organizations should put any model core through their security testing and inspection processes, and they will also often post-train the models for their specific uses, and can examine areas like bias, toxicity, hallucinations, sensitivity to certain topics. So they work with, optimize, and understand the models before people start sending them prompts.

Could a model that is used for coding somehow throw malicious backdoors into the code it writes? Again, while that’s theoretically possible, it would require acrobatic feats to accomplish.

“There’s no reason that a sophisticated enough actor couldn’t train a model to be a completely amazing coding model in every circumstance, but when presented with a certain type of code base … some hidden training would kick in,” Atkins, who spends his days training models, postulated. But he adds: “I don’t know how you would do this.”

Because large language models are by nature creative, the odds are slim of getting a contemporary model to spit out malware in response to a preplanned perfect storm of context and prompt. Even slimmer are the chances that any enterprise would then use that code.

Could it happen in the future? That’s anyone’s guess. But enterprises are also building their AI apps to be model-agnostic and to use multiple models. So even if Chinese models are the best for the price today, enterprises won’t be locked into using them forever.

“I think instead of the conversation being about how to ban Chinese models, it should be about how do we foster a good, open ecosystem here in the U.S.,” Atkins says.

Arcee also gains advantages from Chinese models. Because they are open, the startup “benefits from those models being good because we can learn what they did. We can build on top of them. Then they can learn what we do,” he says. “We have tremendous respect for the people building those models, the individual researchers.”

Ultimately, the way to compete with Chinese models “is to release a model that is better,” says Atkins. “We need to give them something to talk about.”

OpenAI announced Wednesday that it will spend $750 billion on infrastructure through 2030, some 25% more than it estimated earlier this year, the Wall Street Journal reported The renewed blitz comes as its Stargate data center project appears to have stalled.

The first salvo in OpenAI’s spending spree will be a $20 billion data center campus in Georgia known as Project Camellia. The development will span 1,400 acres northwest of Savannah and will draw at least 3.2 gigawatts of power from Georgia Power, the region’s utility. The generating capacity is expected to become available between 2028 and 2032.

The AI company said it would “pay the full cost of the infrastructure and electric-service costs” for the new data center. The Georgia Public Service Commission (PSC) adopted a rule last year to prevent utilities from passing on costs associated with new users drawing more than 100 megawatts. 

Georgia Power also said OpenAI will reduce its power draw by up to 1 gigawatt during periods of high demand on the grid.

OpenAI is receiving a 50% property tax abatement for 15 years from Effingham County, according to the Effingham Herald.

Neither OpenAI nor Georgia Power has said how Project Camellia will be powered. TechCrunch asked both companies for specifics, but did not immediately receive a reply.

Regulatory filings might provide some clues. In December, Georgia Power received approval from the PSC to produce an additional 9,885 megawatts. The utility told the PSC it expects to have all the capacity contracted by the end of 2026. The OpenAI deal accounts for about a third of that.

Based on documents Georgia Power filed with the PSC, most of the new capacity will come from natural gas. The utility said it will build or buy from third parties about 5.8 gigawatts of natural gas generating capacity, about a quarter of which will be from more polluting simple-cycle turbines. Altogether, the new fossil fuel capacity will more than double Georgia Power’s natural gas fleet. The remainder will be supplied by grid-scale batteries and solar.

While electricity from Georgia Power is expected to start flowing in 2028, OpenAI did not give a timeline for when the first GPU will be turned on. That could happen sooner than 2028 given that OpenAI recently hired Brett Mayo to lead data center construction. Mayo previously worked at xAI, where he oversaw the Colossus data center in Memphis. 

Colossus was built in record time, but it has allegedly taken its toll on local air quality, according to a lawsuit filed by the NAACP and the Southern Environmental Law Center. The xAI data center has been running dozens of unpermitted natural gas turbines, claiming exemption from federal clean air regulations.